University
stringclasses 19
values | Text
stringlengths 458
20.7k
|
|---|---|
Chalmers University of Technology | 4. Process mining
Figure 4.16: The process model of UI.
As shown in the both figures, the green points are the start place T , and the red one
I
is the end place T . The blue boxes are the transitions, and the deeper the color,
O
the higher frequency they have. Although the two models are enough for further
research (for example, machine learning), one cannot intuitively find the difference
betweentwomodels. Ifwefilterthesmallprobabilityeventsandarcs, moreintuitive
models which can represent the main flow of the events hidden behind the rather
complex processes of figures 4.15 and 4.16, as illustrated in Figure 4.17. Compared
to the BASE, participants would do something (release the gas pedal) before the
emergency event happen. This confirmed that 3DATIS do made the participants’
behavior change, and the most important was that it made participants behaviors
safer, i.e. they slow down before the dangerous event happened. Furthermore, with
3DATIS, the Accelerate happens before turning the steering wheel. On the con-
trary, Accelerate happens after turning the steering wheel. It is probably because
with 3DATIS, the speed has already slowed down, drivers need to accelerate in order
to overtake the obstacle whereas without 3DATIS, drivers need to avoid the obstacle
first, and then overtake.
43 |
Chalmers University of Technology | 5
Conclusion
In this thesis work, three main tasks for 3DATIS have been done: design, test and
analysis. In the design task, the main question was solved: how to present the
sound in order to make drivers easily understand the surrounding traffic informa-
tion. Then, we tested 3DATIS with 30 participants and each of them did two rounds
of experiments, one with 3DATIS and one without. Two groups of behavior data
were collected from the experiments. When it came to data analysis, the driv-
ing behavior of each scenario was studied. The performance of 3DATIS in Cutin,
Pedestrian, Overtake and Intersection was fairly good. Especially in Intersection
and Pedestrian, 3DATIS gave participants very positive indications to reduce the
rate of collision and improve the TTC. However, in Redcab, except TTC (Time to
collision), all the other indices went in the wrong direction. Most participants were
not prepared and performed in panic with 3DATIS. As there were two obstacles in
this scenario and one of them was the disturbance, participants could not distin-
guish where the sound came from and thus their judgements were interfered. Also
according to the questionnaire and interview, 3DATIS was sometimes annoy when
there are more than one other road user in the road at the same time. Therefore,
3DATIS can hardly exert its advantages in a relatively sophisticated traffic envi-
ronment and even disturbed participants’ judgement. In Chapter 4, a rather recent
method called process mining to discover the process behind a given data set was
introduced. Through process mining, the process models of scenario Cutin was dis-
covered. By comparing the models with and without 3DATIS, the driving behavior
with the help of 3DATIS was found to make participants behaviors safer.
The limitation of this thesis work is mainly focused on the experiment. First, there
were 30 participants that took part in the experiment and this number is almost
the baseline that an experiment requires. With more participants, the conclusion
would be more robust and reliable. Secondly, as each participant was required to
finish two experiments with the same scenarios, the learning effect cannot be totally
avoided, even though the randomization of scenarios can theoretically eliminate the
bias. Thirdly, the steering wheel and pedals of the simulator was designed for the
game, so it could make participants feel not like driving in a real world. The con-
sequence of which is that collected data might not reflect the real driving habit of
participants.
45 |
Chalmers University of Technology | Sensor calibration for autonomous mine vehicles
Calibration of laser scanners and hinge angle sensor
MAGNUS OLOFSSON
Department of Applied Information Technology
Chalmers University of Technology
Summary
Atlas Copco has delivered mining products since the last century and in 2007 they
introduced an autonomous centre articulated LHD (Load-Haul-Dumper). LHD:s are
used in most underground mines for transport of ore and the idea of an autonomous
LHD was to improve safety, efficiency and productivity in underground mines. The
autonomous LHD contains a number of sensors including an odometer, a hinge angle
sensor in the articulation joint, a gyro and two laser scanners needed for autonomous
operation. All of them need to be calibrated because of pose errors from high tolerances
in the machine construction.
In this thesis a method for calibrating the angular offset of the hinge angle sensor and
another method for calibrating the angular offsets of the two laser scanners using data
from the already existing sensors are presented. The hinge angle method relies on the
gyro, the odometer and a kinematic model of a centre articulated vehicle to estimate the
offset. For calibration of the laser scanners the offsets are augmented on to the states in
the state space model used for positioning of the LHD. The augmented state is then
estimated using a combination of an Extended Kalman filter and an Unscented Kalman
filter in a SLAM algorithm that uses the kinematic model and the laser measurements.
Experiments in simulated and real environments have shown that the hinge angle sensor
estimates the angular offset to within 0.3o. Tests of the laser scanner calibration method
have shown that it estimates the angular offset to within 0.5o. The methods run offline,
are easy to operate and require that the operator drives at least 50 m in a straight mine
drift at 2.5 m/s to collect enough data.
The report is written in English.
Keywords: calibration, offset, LHD, centre articulated, laser scanner, hinge angle,
SLAM, Kalman filter.
iii |
Chalmers University of Technology | 1 Introduction
1.1 History
The mining industry is a dangerous workplace, especially underground mining.
During the last decades mining companies have tried to improve efficiency,
productivity and safety. Atlas Copco AB started out in 1873 as a railway
company but turned into air compressors and air tools in the end of the 19th
century. Today they are in a number of markets and at the beginning of the 20th
century they started producing air driven rock drills to the mining industry. The
development has moved on and today they produce a number of mining
products including the centre articulated LHD (Load-Haul- Dump) Scooptram
ST14 for loading, hauling and dumping materials in underground mines.
All LHD:s have traditionally been manually driven by an operator placed in the
machine. But for some years ago solutions began to be used where the LHD
was remote controlled and teleoperated from a control room (Hainsworth
2001). With teleoperation the operator was removed from the dangerous mine,
but teleoperation includes delays and the cameras restrict the view for the
operator which decreases the productivity (Larsson, Broxvall & Saffiotti 2010).
To solve the problems related to teleoperation and also remove the need of a
operator to each LHD autonomous systems have been developed and are
commercially available, see (Automine 2011), (Scooptram Automation 2011)
and (MINEGEM 2011). Atlas Copco’s Scooptram Automation includes both
teleoperation and autonomous operation.
1.2 System overview
Atlas Copco’s autonomous system works using the principle Teaching, Route
profiling and Playback. First the LHD is driven manually along the route it
should operate on and all sensor data are logged. During route profiling data are
converted to a number of grid maps (Marshall, Barfoot, & Larsson 2008) and a
route profile containing waypoints, desired speeds and pause points for loading
and dumping.
During playback the route is followed with help of the sensors. Data from a
hinge angle sensor in the articulation joint, a gyro and an odometer are
weighted together using an EKF (Extended Kalman Filter) to estimate the
position. Laser scans from two laser scanners are then compared with the map
using UKF (Unscented Kalman Filter) to update the position estimate.
Deviation from the desired route point is used as input to a control algorithm
that steers the LHD towards the route point in the desired speed. The system is
built to operate in an obstacle free environment, but simple obstacle avoidance
algorithms are built in the system to make the machine stop if obstacles come in
its way.
1 |
Chalmers University of Technology | 1.3 Objective
All of the sensors need calibration because of errors from high tolerances in the
machine construction. This project focuses on calibration of the hinge angle
sensor and the laser scanners. The hinge angle sensor is used to measure the
angle between the front and rear part. A method was searched that could find an
angle offset that when applied reduces it to within ±0.2o. Before this project it
was required that someone climbed on the back of the LHD and tried to align
the machines back- and front end while the LHD was moving. When aligned
one was able to get a reading of the angular offset.
To get correct maps it’s important that the orientation of laser scanners relative
the LHD:s coordinate system is known. This project is focused on the laser
heading, which is the most critical parameter. The goal was to find the angular
offsets of the laser scanners to give a maximum scan matching error less than a
predefined value of 0.35. Before this project the lasers heading where manually
put to zero by eye sighting.
1.4 Contribution
A new method is here proposed that can be used on any centre articulated
vehicle to automatically find the offset of the hinge angle sensor. It uses the
kinematic model in (Ridely & Corke 2001) together with the hinge angle
sensor, the gyro and the odometer to get an estimate of the hinge angle offset.
Error propagation was made to get an early estimate of the accuracy of the
method. Different confidence intervals were then estimated to be able to
guarantee its performance. Experiments have shown that the method delivers a
hinge angle offset with an accuracy of ±0.3o. The requirement is that the
machine is driven straight at least 33 m at 2.5 m/s
A new method is here proposed that can be used to find the angular offset of a
number of lasers on a moving vehicle automatically. It uses an augmentation of
the original vehicle model, an EKF for prediction, an UKF for filtering and a
SLAM (Simultanoues Localization and Mapping) algorithm. No error
propagation nor a confidence interval were calculated for the laser offset
because the method was too complex and too few data existed to get a good
enough confidence interval. Experiments have verified that if the vehicle is
driven straight in a 23 m normal sized mine drift it can estimate the angular
offsets of two lasers to within ±0.5o, which is need to get a maximum scan
matching error less than a predefined value of 0.35.
1.5 Report structure
In Chapter 2 the report evaluates different ideas before going through the
sensors, algorithms and methods used for finding the offsets of both the lasers
and the hinge angle sensor in Chapter 3. After going through the methods it
continues with dealing with how the driving and sensor measurements affect
the accuracy of the estimated hinge angle in Chapter 4. It then goes through the
real experiments done to verify the accuracy of the system and its results in
Chapter 5. The report then proceeds with taking up the required angular
2 |
Chalmers University of Technology | 2.2 Laser offsets
A literature research showed that there didn’t exist any earlier work in the area
of estimating the laser scanners angular offset. The closest work were the
constant model error estimation in (Martinelli, Tomatis, Tapus, & Siegwart
2003) and (Madhavan, Dissanayake & Durrant-Whytel 1998) which uses EKF
on laser data together with augmentation of the states.
When estimating the laser offsets the laser measurements needs to be used,
which increases the level of complexity compared to the hinge angle calibration
case. One alternative was to by brute force test all positive combinations of
laser offsets to see which two gave the smallest scan matching error but early
experiments showed that it took too long time. The system is built around an
EKF together with an UKF so a natural choice was to use them together with
state augmentation to estimate the laser offset parameters.
A problem with this solution was that the map needs to be known a priori,
which isn’t always the case. To solve this problem we needed to implement
SLAM. UKF based SLAM can be seen in (Martinez-Cantin & Castellanos
2005), which is one of many that uses one state for each feature. The original
system uses a grid map and thereby the feature space become to large if each
grid point should be represented by one state variable. A solution to this was
created in (Merwe, Doucet, Freitas & Wan 2000) using a more complex particle
filter. Instead of using it, it was concluded that one could create a less complex
algorithm that reused the existing mapping algorithm together with the EKF
and UKF. The computations of this method are still heavy and depend on a
number of parameters but its faster than brute force search.
3 Theory and methods
This chapter goes through the theory and methods used for estimating the hinge
angle offset and the laser scanners angular offset. The chapter begins with
going through the kinematic model of the centre articulated LHD in chapter 3.1
and then continues with the sensors used in Chapter 3.2. Chapter 3.3 goes
through the hinge angle offset estimation and its error analysis together with the
statistics used for calculation of the confidence intervals. Chapter 3.4 then
continues with estimation of the laser scanners offset using EKF and UKF in a
SLAM algorithm. The chapter ends with a brief walk-through of the
implementation of the two methods and also the simulator used for creating
simulated log-files.
4 |
Chalmers University of Technology | 3.1 Kinematic model
l
F
l
R
Figure 1: Model of a centre-articulated vehicle.
The kinematic model of the LHD already in use in the system assumes no slip
and can be found in a number of papers (Altafini 1999) and (Ridely & Corke
2001). It uses the setup seen in Figure 1 where g is the angle between the
centre lines of the two bodies. q , q is the heading of the front and rear body
R
respectively in relation to the x-axis of the global coordinate system. l is the
F
distance between the hinge G and the front axle centre F and l the distance
R
between the hinge G and the rear axle centre R. The models for the rotation of
the LHD are
&
vsing +l g& ( 1 )
q = R
l cosg +l
F R
and
& vsing - l g& cosg ( 2 )
q = F
R l cosg +l
F R
where v is the speed of the front body. From the model it can be concluded that
the rotation increases in magnitude with speed, hinge angle and if the hinge is
rotatedg& „ 0. The rotation also decreases with increased machine length.
The LHD has four wheel drive. There are separate differentials for the front and
rear wheels but a stiff driveshaft connects between the two differentials. This
makes the front and back wheels get the same speed. According to the model
they shouldn’t have the same speed if l „ l or the vehicle is not driving
F R
straight,g „ 0, and thereby is a slippage introduced when g is increased. So the
models performance decreases with increased g .
In Ridely & Corke (2001) one error was found in Eq. ( 2 ) where they had
missed the multiplication with cosg , which was proven by simple calculations
and later confirmed by reading Altafini (1999).
5 |
Chalmers University of Technology | The nonlinear discrete time state space model in Marshall, Barfoot, & Larsson
(2008) is created from a continuous one using Euler steps giving
s(n+1) =Φ s(n)+Γ (s(n))(cid:215) (u(n)+v(n)), ( 3 )
s u
x(n)
y(n) v(n)
s(n) = , u(n) =
q (n) w (n)
g(n)
where
hcosq (n) 0
hsinq (n) 0
Γ (s(n)) = - hsing(n) - hl
u R
l cosg(n)+l l cosg(n)+l
F R F R
0 h
v(n) is model errors and other uncertainties and assumed to be Gaussian noise
v(n)~ Ν(0,Q)
with covariance
s 2 0
Q = v
0 s 2
w
h in the model is equal to the sampling time and w (n) is equal to the angular
speed of the hinge angleg& (n). The state update matrix Φ is the identity matrix
s
because we don’t model the dynamics.
3.2 Sensors
Figure 2: The LHD and its sensors.
As seen in Figure 2 the LHD is equipped with a hinge angle sensor to measure
the angle between the front and rear part. It uses and odometer mounted on the
driveshaft in order to measure the speed and the distance travelled. Mounted on
the top of the machine are two laser scanners used to map the surrounding and
position the LHD. One of the scanners is mounted backwards on the rear part
6 |
Chalmers University of Technology | and one is mounted forward on the front part. Together with the front laser an
IMU (Inertial Measurement Unit) is mounted, containing a heading gyro to
measure the rotation of the front part.
3.2.1 IMU Navigation02
The IMU Navigation02 is delivered by the company AIMS. It has three gyros
with 24 bits resolution each giving angular speed resolution of 16m o /s. The
following parameters can be found for the gyros:
• Range ±120 o/s
• Bias Error 0.06 o/s (1 s )
• Scale Factor Error 0.25 % (1 s )
• Non-linearity 0.5 % of FS
• Noise 0.1 o/s (Broadband RMS)
• Bandwidth 25 Hz
• Misalignment 5 mrad
It’s only the heading gyro that is used and a simple model of its measurements
is
& & &
q (n) =q (n)+q +v (n)
m b q&
m
where
v (n) ~ N(0,s 2 (n))
q& q&
m m
and
& &
s (n) =120*0.005/3+0.0025q (n)+0.001=0.3%+0.0025q (n).
q&
m
& &
q (n) is the true angular speed, q is the bias error and v (n) is the Gaussian
b q&
m
noise affecting the measurements. The non-linearity and the scale factor are
modelled as Gaussian noise even though they are constant for each unique IMU
unit. Reasons for modelling them as Gaussian noise are that they aren’t known
&
and that they vary with the size of q so Gaussian noise is the closest
b
approximation. It’s approximated that three standard deviations contains all
errors and that is why the non-linearity is divided by three to get one standard
&
deviation. The bias q is instead modelled as a constant because of that it’s
b
later estimated and removed in the analysis and its sample variance is small.
Misalignment is approximated to be zero due to its small size. It should also be
mentioned that the IMU is mounted in a rubber suspension to remove some
vibrations.
7 |
Chalmers University of Technology | 3.2.2 Odometer
The odometer is base on a driveshaft encoder that measures the number of teeth
passing by to calculate the distance travelled. It uses the fact that if the distance
between the teeth, the radius of the wheels and the number of teeth passed are
known the distance travelled d can be determined. The mounting on the
m
driveshaft makes the odometer insensible to turns.
The odometer system consists of three parts, an encoder, an I/O module and the
machine computer, which is illustrated in Figure 3. The I/O module checks the
encoder pulse counter every 5 ms and the machine computer then checks the
I/O module around every system sampling time h, equal with 40 ms. This
makes it possible that the value in the I/O module is up to 5 ms old. When
calculating the distance travelled the system calculate it as 40 ms has gone
between two samples, which then isn’t always true. The error can never grow
larger than a time error of around 5 ms.
Machine computer I/O module Encoder
40 ms 5 ms pulse
counter
Figure 3: Odometer sample flow from sensor to machine computer
The measured speed v is estimated from d using the model
m m
d (n)- d (n- 1) v(n)h– v(n)*5*10- 3 v(n)*5*10- 3
v (n) = m m = =v(n)– .
m h h h
It’s approximated that the synchronization error in each sample is zero mean
Gaussian. According to the Gaussian distribution 3s » max error which
v
m
makes
v(n)*5*10- 3
s (n) = .
v m 3h
Drift due to slip is approximated to be zero according to previous
measurements done with the machine. Also the radius of the wheels is said to
be known without errors.
3.2.3 Hinge angle sensor
The hinge angle sensor is an encoder mounted in the joint holding the front and
rear body together. The encoders range is (0,360]o and it has 16 bits resolution
giving an angular resolution of 6 millidegrees. Previous measures have shown
that the angular measurements g (n) have a Gaussian noise v (n), which
m g
m
have a standard deviation s of approximately 0.01o. The hinge angle sensor
g
m
is mounted to give approximately 180o when the machine is straight. 0o is
wanted so the offset g is 180o under ideal conditions but there is also a
o
mounting error that introduces an offset. The machine can operate in the range
8 |
Chalmers University of Technology | [-45,45]o where 0o is when the machine is straight. A measurement from the
hinge angle can be modelled as
g (n) =g(n)+g +v (n) ( 4 )
m o g
m
where g(n) is the true hinge angle and
v (n)~ N(0,s 2 ).
g g
m m
3.2.4 SICK Laser scanner
The SICK laser uses a moving laser beam to scan the environment in one plane.
It has a maximum scanning range of 32 m in the interval [-90, 90]o degrees with
an angular resolution of 1o giving N =181 scans. A laser beam i where
points
i=1,…, N can be modelled as
points
z (i) = z (i)+ z +v (i)
m spot ci z
where
v (i) ~ N(0,s 2).
z z
z is a constant error in the range [0,0.03] m and s =0.01 m is the stochastic
ci z
error. The laser beam creates a spot, with aperture angle 0.11 milliradians, that
increases with distance and angle to the objects normal. z (i) is the distance to
spot
the closest point on the object hit by the laser beam.
3.3 Hinge angle offset
3.3.1 Estimation of the hinge angle offset g
o
The idea is to rearrange Eq. ( 1 ) in Chapter 3.1 to get an expression for the
hinge angle and then use it for estimation of the hinge angle offset g .
o
ˆ&
Estimation is made by using the unbiased rotation speed q (n) measured by
ubm
the heading gyro, the measured speed v (n) and the hinge angle
m
measurementg (n), where n=1,…,N is the number of measurements. To
m
simplify the rearrangement estimation is done only for the rotation created from
forward motion and not from change in hinge
angle.g&
is therefore put to zero
and Eq. ( 1 ) is rearranged to give
& &
vsing - q (cid:215) l cosg =q (cid:215) l .
F R
It is known that adding two sinusoidal waves with the same frequency and
different phase gives a new wave with the same frequency but a new phase and
amplitude. For example
asinx+bcosx = a2 +b2 sin(x+j )
9 |
Chalmers University of Technology | where
b 0 if a ‡ 0
j = tan- 1 +
a p if a <0
Applying this makes it possible to get only oneg in the equation
v2 +( q& l ) 2 sin tan- 1- q& l F +g =q& l .
F R v R
Solving for g gives
& & ( 5 )
ql ql
g =sin- 1 R +tan- 1 F
( )
v2 + q& l 2 v
R
and for negative v
& & ( 6 )
ql ql
g =- sin- 1 R ( ) +tan- 1 F .
v2 + q& l 2 v
R
&ˆ
The rotation q (n) includes both the rotation caused by the speed v(n)and by
ubm
g&
(n). To remove the rotation caused by
g&
(n) we use the approximation
g (n)- g (n- 1)
gˆ& (n) » m m
h
which is independent of g and thereby only contains the articulation rate plus
o
some noise. The approximation is used in the non-speed part (v equal zero) of
&ˆ
Eq. ( 1 ) and is then subtracted from q (n) giving
ubm
& &ˆ l
g&ˆ
(n)
( 7 )
q (n) =q (n)- R .
v ubm l cosgˆ(n)+l
F R
where gˆ(n) is an estimate of the true hinge angle.
If g (n) is used as gˆ(n) it introduces an error in the final offset estimate
m
gˆ because of the offset. The error introduced is limited as long as g is
o o
reasonably close to zero but that isn’t always the case because the offset g can
o
be between (-180, 180]o. From the analysis in Chapter 4.2 it has been found that
if the whole algorithm is iterated m=1,…,M times and gˆ(n) is estimated from
g (n) by removing the previous offset estimate gˆ (m-1) the offset estimate
m o
gˆ (m) converges towards the true offset g . Given that the vehicle is driven
o o
along a straight line a first estimate of gˆ can be calculated from
o
10 |
Chalmers University of Technology | 3.3.3 Error propagation
Using error propagation the propagation of the stochastic errors through the
&ˆ
hinge angle method can be analysed. By inserting combinations of q (n),
ubm
v (n), g (n) and the number of measurements N the variance of gˆ can be
m m o
found. It should be remembered that this requires that the model is correct and
thereby doesn’t the analysis give any information regarding effects from errors
in the model.
In Eq. ( 7 ) the measurement error in all parameters are independent. The error
propagation through Eq. ( 7 ) can then be estimated using
& & &
¶ q ¶ q ¶ q
s 2 = v s 2 + v s 2 + v s 2
q& &ˆ q& ¶ g g ¶ g g
v ¶ q m m m
m m- 1
ubm
where
&
¶ q
v =1,
&ˆ
¶ q
ubm
& ( )
¶ q l +l cosg +l g - g sing
v = - l h 2 1 m 1 m m- 1 m
¶ g 2 ( l +l cosg ) 2
m 2 1 m
and
&
¶ q l
v = ( 2 ).
¶ g h l +l cosg
m- 1 2 1 m
In Eq. ( 5 ) the measurement errors are also independent giving
¶ g ¶ g
s 2 = q& ,v s 2 + q& ,v s 2
g q&,v ¶ q& q& v ¶ v v m
v m
where
¶ g q&
,v =
1 l
2
v2 +q& v2l 12 - ( v2 +q& v2l 12) - 1 2l 12l 2q& v2
+
l
1
¶ q&
v 1-
q& v2l 22 v2 +q& v2l 12 v+q& v2l 12
&
v2 +q 2l2 v
v 1
and
¶ g q&
,v =
- 1 q& vl 2v
-
l 1q&
v .
¶ v
m 1-
q& v2l 22 (
v2
+q& 2l2) 23 v2 +q& v2l 12
& v 1
v2 +q 2l2
v 1
12 |
Chalmers University of Technology | The above equations holds for each measurement n=1,…,N. The total variance
of the estimated offset gˆ is then from Eq. ( 10 )
o
(( ) )
1 N ( 11 )
sˆ2 = ∑ s 2 (n)+s 2 (n) +2r (n)s 2 (n)s 2 (n) .
gˆ o N2
n=1
g q&,v g m g q&,v,g m g q&,v g m
g is in Eq. ( 7 ) and in Eq. ( 9 ) which implies that there is a correlation
m
r . The variance s 2 is largest when r =1 but experiments have
g q&,v,g
m
gˆ
o
g q&,v,g
m
shown that the effect of the correlation is small.
Using sˆ one could estimate a 99 % confidence interval for the gˆ . The
gˆ o
o
sensors have Gaussian errors and therefore the error on gˆ is also Gaussian.
o
Using the Gaussian distribution the confidence interval is given by
gˆ - 2.576sˆ <g <gˆ +2.576sˆ
o gˆ o o gˆ
o o
3.3.4 Measurement statistics
By repeating the same experiment N times and each time collect N
1 2
measurements the 99 % confidence interval of the hinge angle method can be
found. There exist two methods to calculate the confidence interval.
The first method begins with calculating the estimated standard deviation
s (p), where p=1,…,N , of the sample offsets gˆ (n) using
gˆ 1 so
so
s (p) = 1 ∑N 2 ( gˆ (p)- gˆ (n)) 2
gˆ so N - 1 o so
2 n=1
where n=1,…,N is the number of samples in each experiment. If N is large an
2 2
approximation can be made that s (p) is equal to the true standard deviation
gˆ
so
s (p)of the whole population and thereby having a Gaussian distributed and
gˆ
so
not a Student-t distribution.
We assume that the spread of gˆ (p) is not correlated as a result of that the
o
measurement noise can be assumed to be white Gaussian noise. Then the
estimated standard deviation of the gˆ (p) from all N measurements is given by
o 2
1 ( 12 )
s (p)= s (p).
gˆ 1 gˆ
o N so
1
According to the Gaussian distribution one could know with 99 % confidence
that the true offset g is in the interval
o
gˆ - 2.576s (p)<g <gˆ +2.576s (p).
o gˆ 1 o o gˆ 1
o o
This should be tested for all N experiments. One experiment had been enough
1
in the ideal world to assure the statistics but more experiments with the same
13 |
Chalmers University of Technology | setup are made to confirm the result. The interval should be smaller than ±0.2o
to confirm that the method is good enough. Ideally it should be enough to
increase N to get into that interval. Also if the spread is correlated s (p) is
2 gˆ 1
o
going to be smaller than the true standard deviation.
The second method calculates the standard deviation of gˆ (p) directly. To begin
o
with the mean of all gˆ (p) is
o
gˆ = 1 ∑N 1 gˆ (p)
o N o
1 p=1
and the estimated standard deviation is then
s = 1 ∑N 1 ( gˆ (p)- gˆ ) 2 .
gˆ o N - 1 o o
1 p=1
Due to that N is relatively small s has an uncertainty. If the spread is
1 gˆ
o
Gaussian, s has a c 2 distribution and the 99 % confidence interval of s is
gˆ gˆ
o o
(N - 1)s2 (N - 1)s2
L = 1 gˆ o <s < L = 1 gˆ o .
1 c 2 gˆ o2 2 c 2
0.99/2,N - 1 1- 0.99/2,N - 1
1 1
The confidence interval of g is then
o
gˆ - 2.576L <g <gˆ +2.576L
o 2 o o 2
This method only relies on that the spread is Gaussian and is therefore more
reliable. It gives a larger confidence interval due to the uncertainty in s . The
gˆ 2
o
uncertainty can be decreased by increasing N . When N goes to infinity the
1 1
confidence interval decreases to the true confidence interval.
One could compare s with each of the N estimated standard deviations s
gˆ 1 gˆ 1
o o
by assuming that s is the true standard deviation and using the fact that s is
gˆ gˆ
o o
the spread of the gˆ (p):s around gˆ . This gives
o o
c 2 = 1 ∑N 1 1 (gˆ (p)- gˆ )2. ( 13 )
N - 1 N - 1 s 2 o o
1 1 p=1 gˆo1
c 2
If >1 than s <s and we have under estimated the standard deviation
N - 1 gˆ o1 gˆ o
1
using the first method. This means that the spread is probably not uncorrelated.
c 2
can be use to scale s to its true value but it should be remembered
N - 1 gˆ o1
1
that Eq. ( 13 ) doesn’t take in to consideration that s is an uncertain estimate
gˆ
o
of s if N is small.
gˆ 1
o
14 |
Chalmers University of Technology | 3.4 Laser scanners angular offset
3.4.1 Creating and evaluating the map
Under normal operation is the map M created under Route Profiling from the
log data, containing all the sensor data recorded under the teaching step. To
create M the log is gone through step by step to estimated u(n) using backward
Euler ong and d . u(n) is then input to Eq. ( 3 ) to get the pose s(n) of the
m m
machine at each time step. The map, which is represented by a 2 dimensional
binary grid is then finally updated using ray tracing along the recorded laser
beams at each position s(n) for both the front and rear laser.
After creation a simulated Playback is made in the map to evaluate it using the
logged data as input. In the simulation the logged laser beams’ lengths are
compared to the simulated ones. From the comparison over all samples a scan
matching error e is calculated. If the map is identical to the real world the
match
machine should get identical laser scans and e equal to zero. The e
match match
must at all times be under 0.35 for the map to be acknowledged.
3.4.2 State estimation using nonlinear Kalman filter
To keep track of the machine under playback the state s(n) needs to be
estimated at each sample time. A good prediction of the state can be found by
using the kinematic state space model in Eq. ( 3 ). To get better accuracy and
compensate for drift the state is then filtered with help of the measured laser
data. This method of first predicting using a model and then update using
measured data is called Kalman Filtering. A standard Kalman Filter can only
operate on linear equations. For nonlinear equations there exist a number of
different solutions. This machine uses a mix of two Kalman Filters, an EKF
(Extended Kalman Filter) for prediction and an UKF (Unscented Kalman
Filter) for filtering. The implementation of the EKF and UKF used can be seen
in (Marshall, Barfoot, & Larsson 2008). To simplify notation let sˆ- (n) be the
prediction of the state using EKF and sˆ(n) the filtered estimate of the state
using UKF. In the same way let P- (n) be the predicted state covariance and
P(n)be the filtered covariance.
To solve the problem that the Kalman filter needs linear equations the Extended
Kalman Filter uses linearization of the state equation Eq. ( 3 ) around the
previous state. Eq. ( 3 ) is linearized to get
sˆ- (n) = H (sˆ(n- 1))(cid:215) sˆ(n- 1)+Γ (sˆ(n- 1))(cid:215) u(n). ( 14 )
s u
15 |
Chalmers University of Technology | s 2
xinit
s 2 0
yinit
s 2
P(0)= q init .
s 2
g
0 init s 2
f
Rinit
s 2
f
Finit
3.4.4 SLAM
The UKF needs a map to be able to estimate f and f but the mapping
R F
procedure in Chapter 3.4.1 is affected by f and f because of that can’t the
R F
map be created before the offsets are estimated. It needs to be done
simultaneously.
The map M could be created by adding laser scan information z(n) to M at each
sample n using sˆ (n), estimated from Kalman filtering using artificial
a
~
measurements Z(n) on M at n-1. This method has some drawbacks, one being
~
that earlier estimates of f and f affects Z(n) and experiments showed that
R F
this made the Kalman filter unable to make f and f converge to their true
R F
values. It’s solved by recreating the map around the machine at sample n from
~
scratch for each setup of f and f in S(n). To recreate the map ray tracing is
R F
used in each old pose in sˆ (n+n ) with scan z(n+n ), where
a minus minus
n = -n -2n ,…, -N and n is the updating step used to decrease
minus step, step minus step
computation time.
Another drawback is that the map is less updated in the driving direction. This
is solved by using the EKF to estimate future states sˆ- (n+n ), where
a pozitive
n =n 2n …, N The poses in sˆ- (n+n ) are then used
positive step, step positive. a pozitive
together with the laser scans z(n+n ) to add data to the recreated map for
positive
~
each sigma setup in S(n) of f and f . This works over short travelled
R F
distances given that the model is correct.
Each ray tracing to a point on the map overwrites the old value. So the update
order is of importance. The start position of the machine is exactly known when
mapping, because we decide its coordinates. Poses in negative time are also
more accurate because of that they have been estimated using the whole
Kalman filter. For this reason it is more convenient to start with mapping in the
future poses and end with mapping in the negative poses.
To get a more stable final output value the mean of the laser angle estimates
fˆ
R
and fˆ in sˆ (n) from samples at n>N are used to get f and f . The first
F a SM R F
N samples is skipped to give the estimated offsets time to settle because it
SM
19 |
Chalmers University of Technology | 3.5 Implementation
Laser and hinge angle offset estimation is made offline using a log-file
containing all necessary data. The log file is recorded when the vehicle is driven
a certain distance at a certain speed. To be able to create log-files, to test
different laser offsets, an at Atlas Copco already existing SIMULINK simulator
was modified. It simulated the kinematics of the ST14 seen in Eq. ( 3 ), the
sensors and the laser scanning. Laser scans were originally made on maps
created using lines but was modified to also handle grid maps. The laser
simulation was also modified to include the laser spot effect discussed above so
that the final simulator was able to simulate all errors in the laser sensor. Noises
in the other sensors were skipped to decrease the complexity. The grid maps
used were created using Profiling from real logs. The Route Profiling was
together with the Simulated Playback, described in Chapter 3.4.1, already
implemented by Atlas Copco in C++ and ready to use.
The hinge angle offset estimation and error propagation together with the
measurement statistics in Chapter 3.3 were all implemented in MATLAB. As
input to the hinge angle offset estimation method a real log-file was used. A
script was also made to be able to go through a number of log-files to analyse
the confidence interval of the method using the measurement statistics in
Chapter 3.3.4.
All laser offset estimation parts in Chapter 3.4 were implemented in C++ using
Microsoft Visual Studio 2005 and used data in a log-file together with a
configuration file with all parameters as input. The laser offset implementation
was built on modifications of an already existing Route Profiling and Playback
code. From the C++ programs mat-files were delivered that could be analysed
by MATLAB. A script was written in MATLAB to be able to automatically run
estimations from many log-files automatically.
4 Error propagation in the hinge angle method
To see how the errors propagated through the hinge angle offset estimation two
analyses were made using the error propagation in Chapter 3.3.3. The analyses
ˆ&
shows how the unbiased angular speed measurement q (n), speed
ubm
measurement v (n) and hinge angle measurement g (n) affects the standard
m m
deviation of the estimated offset sˆ . A third analysis was made to see the
gˆ
o
effect of just iterating the hinge angle estimation ones using g (n) as the
m
estimated hinge angle gˆ(n) in Eq. ( 7 ).
4.1 Analysis setup
All analyses were made at the systems sample frequency f of 25 Hz. An early
s
analysis shown that the error isn’t affected by g& (n) when smaller than 14 o/s,
m
which is maximum steering speed, so no further analysis for it was made. In all
21 |
Chalmers University of Technology | experiments g (n) was swept from 0 to 45o, where 45o is the maximum steer
m
&ˆ
angle. q (n) was calculated from Eq. ( 1 ) using v (n) and g (n) as inputs.
ubm m m
&ˆ
q (n), v (n) and g (n) were then inserted to the error propagation equations
ubm m m
to give s .
gˆ
o
Different analyses were made starting with investigating how the standard
deviation behaved when only using one sample at different speeds v (n) equal
m
to {1; 2.5; 4.5} m/s. Using more samples N only decrease sˆ with one divided
gˆ
o
by the square root of N and doesn’t change the characteristic.
When travelling at low speed more samples N are collected compared to
travelling at high speed. It was investigated how this affected sˆ using Eq. ( 11
gˆ
o
) at the speeds v (n) equal to {1; 2.5; 4.5} m/s when travelling 50 m.
m
A final analysis was made to evaluate the error in gˆ from using g (n),
o m
including the offset g , as the estimated hinge angle gˆ(n) in the denominator of
o
Eq. ( 7 ) and iterate the hinge angle estimation method only one time. It should
be noticed that the error is a constant error that depends on g and not a random
o
error as those above. Here
g&
(n) affects the error and according to Eq. ( 7 ) does
larger
g&
(n) give larger effects, so
g&
(n) was put to its absolute maximum value
of 14 o/s. The error also increase with decreased speed so to find the maximum
error the speed was put to the lowest speed used around 1.0 m/s.
Increased offset also increases the error according to Eq. ( 7 ). To see the effect
of different offsets g , the set {2, 5, 25, 45}o was tested.
o
4.2 Results
Using only one sample to estimate the offset gives the standard deviation sˆ
gˆ
o
seen in Figure 4 at different hinge angles g (n) and speeds v (n). As seen sˆ
m m gˆ
o
increases with g (n) and decreases with increased speed v (n).
m m
The result for travelling 50 m, giving different number of samples, can be seen
in Figure 5. As seen in the figure sˆ increase less for large angles and low
gˆ
o
speeds compared with Figure 4.
Analyses of the effects of using g (n) as gˆ(n) in Eq. ( 7 ) showed that the error
m
in gˆ increased almost linearly with increased g (n) as seen in Figure 6. It
o m
could also be seen that the error in gˆ is always smaller then the true offset.
o
22 |
Chalmers University of Technology | Figure 6: Error in estimated offset at the known offsets {2, 5, 25, 45}o because
of using g (n) as gˆ(n) in Eq. ( 7 ). It should be observed that the error is
m
always less than the true offset.
4.3 Discussion
From Figure 4 and Figure 5 it can be concluded that the estimated standard
deviation sˆ increases with increased hinge angle g (n) and decreases with
gˆ m
o
increased speed v (n). The reason for increased sˆ with increased g (n) is
m gˆ m
o
because of that uncertainty in speed increases the uncertainty in rotation speed
&
q with increased g (n). If the uncertainty in speed is zero then sˆ is almost
m gˆ
o
&
constant for increased g (n). Increasing v (n) gives increased q when keeping
m m
g (n) constant and thereby better readings. It should also be remembered that
m
according to Eq. ( 11 ) increasing the sample size N by driving a longer distance
decreases sˆ . From Figure 5 it could be concluded that when driving 50 m the
gˆ
o
99 % confidence interval 2.576s is only 0.21o even for low speeds at small
gˆ
o
g (n), which is close to ±0.2o.
m
When analysing the error introduced because of using g (n) as gˆ(n) in Eq. ( 7
m
) it was found that the error increased almost linearly with g (n). It was also
m
found that the error in gˆ was always smaller than the true offset which makes
o
it possible to iterate to find the best estimate gˆ of the true hinge angle. The
o
error is a constant error, which means that taking more samples doesn’t help,
but iterating as described in Chapter 3.3.1 reduces the error to zero.
24 |
Chalmers University of Technology | A conclusion is that because of increased sˆ at high hinge angles it’s expected
gˆ
o
that the machine is going to be driven as straight as possible. When using the
mean of g (n) as an first approximation of the offset this gives a remaining
m
offset in the hinge angle estimate gˆ(n) equal to the mean of the hinge
angleg (n), see Eq. ( 8 ). If g (n) is small the error is even smaller according to
Figure 6 so as long as the machine is not driven with to high g (n) no iteration
is needed.
5 Hinge angle experiments
To test the calibration method a series of experiments were needed. A list of
goals below was created to fully test the systems limits.
• Find the minimum length needed to get good enough precision.
• Check the precision in relation to speed. Find minimum speed.
• Check the precision in relation to turn radius. Find maximum angle.
• Check the deadband of the machine.
• See if stretching the machine in one direction gives better performance.
45 experiments to achieve these goals were put up. In most of them one drove
straight or in an arc. No experiment was made with a large hinge angle offset
because the theoretical analysis above proves that it doesn’t work.
The experiments were made on an ST14 LHD in an old underground mine in
Kvarntorp in Kumla by an experienced driver. The data were then saved
analysed offline in MATLAB. The third goal couldn’t be tested because of
limited space in the mine drift.
5.1 Experiment setup
All experiments were repeated at the speeds {1.2; 2.5; 4.5} m/s because they
are the maximum speeds at each gear and thereby easy to keep constant. In all
cases the driver tried to drive relaxed so normal correction oscillations should
be included. The hinge angle sensor was pre-calibrated to give a reasonable
hinge angle. Then the mean of the sampled hinge angles when driving straight
300 m was calculated, which gave the remaining true hinge angle offset with a
precision of around ±0.05o. The true offset was then used as a reference in the
experiments. To get the gyro bias the machine stood still for 10 s before and
after completing a run through the drift to collect gyro bias data.
The first experiments were made to test the method under normal conditions.
To collect data the machine was driven straight one to two times back and forth
through a 300 m long tunnel at the different speeds. A straight path was chosen
because that should give the smallest errors.
To test the last two goals the machine was driven with the true hinge angles of
{0.2; ±0.7}o at different speeds a distance of 100 m in an 11 m wide tunnel. The
25 |
Chalmers University of Technology | hinge angle of 0.2o was chosen to see if better performance was achieved if the
machine was stretched in one direction. The thought was that if one stretches
the machine in one direction the machine should not oscillate because of the
play in the axles and the hinge. This then should increase the performance of
the method. ±0.7o were chosen to see if there was any play in the hinge angle
sensing and thereby different results in the offset estimation.
The reason to not test with random moves was that the errors arise when the
angle is high and/or speed low. Also the error that comes from
increasing/decreasing the angle is small. If totally random moves had been
made the errors had cancelled out.
Enough with data were collected at each setup to be able to divide it into a
number of partions containing N measurements where N was chosen to 250,
2 , 2
500 and 1000 samples, to test the precision at different lengths. Each
experiment setup was repeated N times to find the confidence interval of the
1
estimated offsets.
A number of statistics were then calculated using the data starting with the
mean gˆ of all N offset estimatesgˆ to see that they were centred on the true
o 1 o
offset. Then the maximum deviation from the true offset d was calculated
gˆ max
o
for each experiment setup. d is the maximum error from our
gˆ max
o
measurements, which is only a small amount of the whole population.
Confidence intervals take the whole population into consideration and are more
assertive statistics. The methods described in Chapter 3.3.3 and 3.3.4 were
therefore used to find different 99 % confidence intervals for gˆ . For each
o
measurement n in experiment p the 99 % confidence interval was calculated
using the error propagation with the measurements as input. The largest
confidence interval z max(sˆ (p)), where z =2.576, was then recorded
0.99 gˆ 0.99
o
for each setup.
The largest and smallest confidence interval z max(s (p)) respectively
0.99 gˆ 1
o
z min(s (p)), using the standard deviation in each experiment p, were
0.99 gˆ 1
o
recorded for each setup. By using the estimated standard deviation s of all the
gˆ
o
gˆ :s, the confidence interval with lower boarder z L and higher boarder
o 0.99 1
z L was calculated and recorded for each setup together with the estimated
0.99 2
c 2
confidence interval z s and the mean of all N ratios for each
0.99 gˆ o 1 N - 1
1
experiment setup.
26 |
Chalmers University of Technology | 5.2 Results
From the mean value, when driving 300 m straight, it was found that the true
remaining hinge angle was 0.31 o ±0.05 o. So the goal for the experiments is to
get hinge angle offsets in the range 0.31 o ±0.20 o, to show that the method can
operate under all conditions.
5.2.1 Driving straight
Table 1: (a) The experiment setups, mean hinge offsets gˆ and maximum
o
deviating offsets d when driving with a true hinge angle g of 0o. (b) The
gˆ max
o
Gaussian 99 % confidence intervals for the experiments using different
methods.
a.
g
Exp. v
Setup [o] Gear [m/s] N 2 f s Distance N 1
gˆ
o
d
gˆ omax
1 0 1 1.3 250 20 16.25 19 0.34 0.36
2 0 2 2.6 250 20 32.50 38 0.30 0.20
3 0 3 4.5 250 20 56.25 38 0.32 0.15
4 0 1 1.3 500 20 32.50 9 0.34 0.20
5 0 2 2.6 500 20 65.00 18 0.30 0.11
6 0 3 4.5 500 20 112.50 18 0.32 0.11
7 0 1 1.3 1000 20 65.00 4 0.34 0.19
8 0 2 2.6 1000 20 130.00 6 0.30 0.05
9 0 3 4.5 1000 20 225.00 6 0.31 0.05
b.
E Sex tp u.
p z 0.99max(sˆgˆ o(p)) z 0.99min(s gˆ o1(p)) z 0.99max(s gˆ o1(p)) z 0.99L 1 z 0.99sgˆ o z 0.99L 2
Nc 12
- 1(p)
1 0.26 0.08 0.20 0.20 0.29 0.49 6.40
2 0.13 0.06 0.19 0.14 0.19 0.26 5.39
3 0.08 0.04 0.19 0.12 0.15 0.21 7.94
4 0.19 0.07 0.11 0.15 0.25 0.60 9.47
5 0.10 0.04 0.13 0.10 0.14 0.23 4.06
6 0.06 0.03 0.12 0.08 0.11 0.19 9.88
7 0.13 0.05 0.07 0.14 0.28 1.79 23.23
8 0.07 0.03 0.06 0.03 0.06 0.24 2.61
9 0.04 0.03 0.06 0.04 0.07 0.25 7.30
The results from driving straight with different speeds and using different
number of samples can be seen in Table 1, for figures see appendix B. As seen,
the estimated offsets gˆ have a mean gˆ very close to 0.31o in all cases. If one
o o
looks at their distributions one finds out that they have Gaussian distributions
which tell us that our assumptions made in Chapter 3.3.3 and 3.3.4 are valid.
From Table 1 it could be seen that for all cases, besides the one when only
using 250 samples and driving with the first gear, d is smaller than or
gˆ max
o
equal to 0.20o.
27 |
Chalmers University of Technology | Using gear 2 and 3 instead of gear 1 largely improves the performance
according to d . Also increasing the distance driven and thereby the sample
gˆ max
o
size decreases d for gear 2 and 3 but only a small amount for gear 1 at
gˆ max
o
larger sample sizes. The difference between using gear 2 and 3 decreases with
increased sample size.
Table 1b shows the positive borders of different 99 % confidence intervals
calculated using the methods in Chapter 3.3.3 and 3.3.4. z L and z L tell
0.99 1 0.99 2
in which interval the true positive confidence interval border is. They only
relies on that the measurements are Gaussian and are therefore the most reliable
ones. z max(sˆ (p)) is the most uncertain statistic. As seen in Table 1b it
0.99 gˆ
o
compared to z L underestimates the confidence interval in most cases.
0.99 1
z s (p) also underestimates the confidence interval sometimes because of
0.99 gˆ 1
o
that z min(s (p)) is smaller than z L for all setups and sometimes leave
0.99 gˆ 1 0.99 1
o
reasonable estimates z max(s (p)), which in most cases are in the interval
0.99 gˆ 1
o
defined by z L and z L . But on average it underestimate the confidence
0.99 1 0.99 2
c 2
interval according to the mean of .
N - 1
1
It can be seen from z s that using confidence intervals instead of d
0.99 gˆ gˆ max
o o
gives similar results. By looking at z L it can be seen that for gear 2 and
0.99 2
gear 3 the error is smaller than 0.3o in all cases if one includes the measurement
error of the true offset. Increasing the distance travelled from 33 m doesn’t
improve z L . To see the spread of the offsets and their confidence intervals
0.99 2
see appendix B.
5.2.2 Small turning
Table 2 (a) The experiment setups, mean hinge offsets gˆ and maximum
o
deviating offsets d when driving with a true hinge angle g of 0.2o. (b) The
gˆ max
o
Gaussian 99 % confidence intervals for the experiments using different
methods
a.
g
Exp.
Setup [o] Gear v [m/s] N 2 f s Distance N 1
gˆ
o
d
gˆ omax
10 0.2 1 1.3 250 25 13.00 12 0.30 0.25
11 0.2 2 2.6 250 25 26.00 12 0.27 0.18
12 0.2 3 4.5 250 25 45.00 12 0.37 0.25
13 0.2 1 1.3 500 25 26.00 4 0.29 0.13
14 0.2 2 2.6 500 25 52.00 4 0.28 0.05
15 0.2 3 4.5 500 25 90.00 4 0.39 0.18
28 |
Chalmers University of Technology | By looking at the gˆ at ±0.7o in Table 3 it’s found that turning positive
o
increases the estimated hinge angle gˆ and turning negative decreases gˆ . It
o o
could also be seen that the deviation from the true hinge angle offset decreases
with speed. If Table 2 and Table 3 are compared it can also be observed that
increased turning also increases the deviation in gˆ . One exception is at gear 3
o
in Table 2 where the deviation is equally large as those at other gears in Table
3. To see the spread of the offsets and their confidence intervals see appendix
B.
5.3 Discussion
From the results when driving straight one can draw the following conclusions:
1. Performance improves with higher speed and longer distance.
2. The difference of using gear 3 and 2 decreases with distances travelled.
3. Driving slower than 2.5 m/s and shorter than 33 m is not recommended.
As seen in Table 1 both z sˆ (p) and z s (p) are underestimating the
0.99 gˆ 0.99 gˆ 1
o o
true confidence interval. One of the reasons for z sˆ (p) to underestimate
0.99 gˆ
o
the confidence interval is that it uses a model, which in reality always contains
model errors. It also only takes into consideration the standard deviations of the
sensors that for the speed and the gyro contain modelling errors. Errors like
slippage and vibrations, which affects the gyro, is also not considered. Even
when z sˆ (p) underestimates the confidence interval it still models the
0.99 gˆ
o
behaviour correctly, when compared to z s in Tables 1, 2 and 3.
0.99 gˆ
o
Driving at 4.5 m/s didn’t give any significant improvement in both the straight
case and when a small turn was made. It gave the same confidence interval,
driving 56 m at 4.5 m/s, as driving 65 m at 2.5 m/s, as seen in Table 1. Driving
slower makes it easier to perform the calibration in small and narrow mines so
2.5 m/s is preferred to 4.5 m/s.
z s (p) only rely on the facts that the measurements are not correlated, that
0.99 gˆ 1
o
the sample size is large and that they have a Gaussian distribution. The later has
been confirmed and the sample size is large. By calculating the correlation it
was found that there exists an oscillation with a period of around 1 s in the
sample offsets. By looking at the gyro readings and the machine when the
machine stops it has been seen that the LHD is swaying for some seconds with
a period of 1 s. From this it can be concluded that the LHD has a swaying
resonance frequency of around 1 s and because of the correlation the variance is
under estimated. The error introduced by the swaying is small if large sample
sizes are used.
The experiments made when turning with a small hinge angle shows that there
is nothing gained in precision by doing it. It’s even so that the precision seems
to decrease according to Table 2. When using 500 samples and gear 2, Table 2,
30 |
Chalmers University of Technology | it seems like that the confidence interval is largely decreased, but looking at
z L one understands that it is possible that it’s underestimated. These results
0.99 2
are also backed up by the mean ratio, which normally is larger than 1. The
reason for the mean offset estimate gˆ to be high at high speeds can’t be
o
explained by other than the occurrence of slippages at high speed when turning.
This is probably also the reason for the increased confidence interval, but it
should be remembered that there exists a uncertainty in gˆ .
o
Another conclusion is that when the turning is increased a deadband is created
according to Table 3. It’s not a normal mechanical deadband because it
shouldn’t increase with turning angle. More likely it appears because of
slippages. It can also be observed from z L and z s in Table 1, 2 and 3
0.99 1 0.99 gˆ
o
that the confidence interval and thereby the spread increases with increased
hinge angle. From this a conclusion can be drawn that driving as straight as
possible is recommended.
It should be noted that the z L to z L intervals are too large and
0.99 1 0.99 2
overlapping, because of too few experiments with each setup, in order to with
99 % confidence assure the above conclusions. But there is a high probability
that they are right. It should be remembered that z L and z L depends on
0.99 1 0.99 2
the number of samples and therefore converges to the true confidence interval
when the number of experiments with the same setup goes to infinity and vice
versa. Without more data it can only be assured that the offset estimation error
is less than z L . A effect of this can be seen in the straight case where it
0.99 2
seems as increasing the distance doesn’t decrease z L . This happens because
0.99 2
the sample size decreases with distance. If one instead looks at the estimated
confidence interval z s it decreases. The problem of too few samples is also
0.99 gˆ
o
one of the factors that drives z L high in the cases when making small turns.
0.99 2
From the above experiments it can be concluded that the method can deliver a
hinge angle offset with a precision better than 0.3o with a 99 % confidence as
long as the machine is driven straight 33 m or longer at a speed of 2.5 m/s. The
goal of 0.2o can’t be achieved following the criterions in Chapter 2.1 but it is
possible, with high probability, that more experiments is going to show that the
true 99 % confidence interval lies lower than the here estimated L :s.
2
Later investigations of the LHD used at data collection in Kvarntorp showed
that a part in the hinge angle sensor was broken. This gave the hinge angle
sensor a play. The load on the hinge angle changes when the machine goes in
different directions which then probably affected its readings because of the
play. This can have altered the results negatively and is probably the reason for
the deadband.
31 |
Chalmers University of Technology | 6 Required laser offset accuracy
For the lasers the required accuracy was specified as a maximum scan matching
error in the map. To be able to specify the corresponding required precision of
the angular offset of the lasers a set of offsets in a number of environments
were tested to find the ones giving the maximum allowed scan matching error
of 0.35.
6.1 Analysis setup
To find the offsets giving the maximum allowed scan matching error a grid of
varying laser offset combinations (f ,f ) was created. It was expected that the
R F
laser offsets never would be larger than ±3o degrees. But we chose to use a grid
with the range [-5 5]o in each direction and a resolution of 1o. The grid was
applied on real log data from six different routes in a mine in the Finnish town
Kemi. Three routes were straight and three routes had a turn. For each grid
point and route the mapping and evaluation procedure described in Chapter
3.4.1 was applied using the offsets in the grid as correction. The reference
offsets were estimated from the three straight routes using the above method
with a resolution of 0.3o and were found to be equal to (-0.8;0.3)o, which then
were represented by the vector φ .
ref
6.2 Results
Figure 7: Highest scan matching error from all routes at each offset
combination. Green crosses and the red shaded area in the middle show the
offsets giving a e smaller than 0.35. The red circle shows the offset φ
match ref
and the red square shows the offsets φ with lowest e .
Goptima match
Figure 7 shows the total maximum of the scan matching errors, when driving
the paths with different offsets. The green crosses and the red shaded area in the
middle show the offset combinations that give a e smaller than 0.35 for all
match
32 |
Chalmers University of Technology | In Figure 8a and 8b the maximum scan matching errors from straight paths
respectively paths with a turn can be observed. Figure 8a from the straight paths
shows that the offsets φ with loweste in the straight cases coincides
SOptima match
with φ . It also shows that the area of allowed offset is larger than in Figure
ref
7, centred around φ and allows the offset to vary ±2o. These results are also
ref
valid if one looks at the three paths separately. It could be seen that Figure 8b
from the turning paths resembles Figure 7 since it has a higher e for almost
match
all angles and thereby dominates Figure 7. It should however be mentioned that
it’s only one path that dominates the other three and decides the shape of Figure
7 and 8b. In all turning cases φ varied but the area of allowed offsets was
TOptima
centred around φ and had the same size as in the straight cases with
ref
exception of the dominating one.
6.3 Discussion
It’s the worst cases that set the limits because the scan match error must always
stay under 0.35. From Figure 7 the conclusion can be drawn that the offsets
can’t be allowed to deviate more than -0.5o from their true value. One also
assumes that one can’t allow them to deviate more than +0.5o. This conclusion
comes from the symmetry in the systems dimensions which should give
symmetric scan matching errors and is confirmed by the straight cases where
the allowed area is centred around φ . More cases had been needed to
true
investigate this further but because of too large computation times as well as
lack of relevant data this was not possible.
Figure 9: Data from one of the turning paths. As seen there exist an offset that
gives a lower maximum scan matching error at the turn than the true offset.
34 |
Chalmers University of Technology | The reason for φ to sometimes deviate from φ in the turning cases can
TOptima ref
be explained by Figure 9 showing data from one of the turning paths. In Figure
9 the blue line is e using φ , the red line is e using φ and the
match ref match TOptima
green line is the hinge angle g in radians. From the figures it can be observed
that e goes high when g goes high. φ gives a better e on average
match ref match
but φ gives the lowest maximum e . One of the reasons for e to
TOptima match match
increase in turns is because of the geometry of the mine drift and the side drift,
which the turn is made into, because e is also increasing when just passing
match
a side drift, which can be seen by looking at e for each sample in the
match
straight cases.
Sometimes there are also walls with ventilation tubes or cable ladders that make
the laser data corrupt and increasee . e also goes down at erroneous
match match
offsets because of above mentioned reasons. As a result of that we didn’t
collect the data ourselves, we can’t know where there were things interfering
with our measurement. Nevertheless, non true offsets have shown to sometimes
give better scan matching errors.
7 Accuracy of the laser offset estimation
To evaluate that the method provide reliable offsets with high enough precision
a set of experiments were performed on real and simulated data.
7.1 Test setup
Table 4: Places, angles used at simulation and number of real tests at each
location.
Simulated (f ;f ) [o] Real [number of tests]
R F
Kemi Straight (0,0), (0,3), (3,3) 10
Kemi Turn (0,0), (0,3), (3,3) 3
Kvarntorp (0,0), (0,3), (3,3) 5
A test schema seen in Table 4 was created with different conditions to test. It
contains tests to see if the method could handle both straight and turning paths
in real and simulated cases. The tests were done using log-files from a mine in
Kemi, which has representable proportions. The simulation tests were made to
see that the method could handle all expected offsets f and f because we
R F
couldn’t do real experiments where we turned the lasers. Routes from the same
area in Kvarntorp were used to test how it handled large mine drifts with many
side drifts. It was also of interest to see how fast it converged in the different
cases and how long driving distance that was needed.
The method shouldn’t need reconfiguration for each environment so in all tests
the same parameter configuration was used. A speed of around 2.5 m/s was
used because the early tests had shown that higher speeds gave the best result
and 2.5 m/s is the maximum usable speed. The hinge angles measurements of
35 |
Chalmers University of Technology | the real data where compensated for sensor offset using the method described in
Chapter 3.3.1. In the simulations the same driving path was used in each
location for all offsets to not affect the results. 4o turns were made in the end of
the Kvarntorp runs to see if the Kalman filter could handle it.
When analysing the real paths the true offsets were known for the Kemi data
from the analysis above and equal to (-0.8;0.3)o. In Kvarntorp the true offsets
weren’t known but constant in all routes. The reason to use real cases was to be
able to analyse the accuracy under real circumstances. In the straight Kemi
cases 10 different log-files had been recorded in parts of two areas in the mine.
As a result of this the following tests are in the same areas
• 1, 3, 4 and 8
• 6, 9 and 2
• 7, 10 and beginning of 5
The Kalman filter uses the scan matching error e to update its estimates
match
and a larger e makes the algorithm to converge faster. As seen in Figure 7
match
equal offsets and opposite offsets give almost the same scan matching error
e . If (f ,f ) are (3,0)o or (0,3)o it should be harder than if they are (3,3)o to
match R F
find them, due to lower scan e . But the convergence shouldn’t depend on if
match
the offsets are (3,0)o or (0,3)o. From this a conclusion was drawn that it’s
enough to test with the offsets (f ,f ) equal to {(0,0) (0,3) (3,3)}o.
R F
7.2 Results
7.2.1 Driving straight in a normal mine drift
f f
R F
Figure 10: f and f from the simulated straight path in Kemi at laser offsets
R F
{(0,0) (0,3) (3,3)}o.
36 |
Chalmers University of Technology | In Figure 11 f , f , fˆ and fˆ can be seen from data recorded along real
R F R F
straight paths in Kemi, see maps in Figure 12. From looking at the figure it can
be seen that the offsets are spread around (f ,f ) equal to (-0.8; 0.3)o, which is
R F
the true offset. 8 of the 10 final (f ,f ):s are inside the ±0.5o limit and the
R F
same 8 are inside the limits after only 23 m. By looking at
(fˆ ,fˆ
) it can be
R F
seen that some estimates converge fast and some takes longer time, especially
for
fˆ
. All of them have converged close to their final value after 40 m and
F
(f ,f ) have get close to their final values after around 50 m. The computation
R F
time was around 17 minutes in all cases.
7.2.2 Turning in a normal mine drift
f f
R F
fˆ fˆ
R F
Figure 12: Angular offsets estimated from a simulated route in Kemi with a
turn and laser offsets {(0,0) (0,3) (3,3)}o. As seen, fˆ get affected at the turn
F
after 55 meters.
Figure 12 shows f , f , fˆ and fˆ when simulating with the offsets {(0,0)
R F R F
(0,3) (3,3)}o. The offsets have converged to within ±0.1o when the mean
valuing is started after 23 m and doesn’t get largely affected by the turn after 50
m. As seen by
(fˆ ,fˆ
) the offsets converge after a few meters.
fˆ
does get
R F F
affect in the turn but then returns to its original value. For
fˆ
it’s hard to see if
R
it is affected by the turn. Its largest reaction happens after 70 m, which is when
the back has left the turn and is facing towards it.
38 |
Chalmers University of Technology | f f
R F
fˆ fˆ
R F
Figure 13: Angular offsets estimated from real data recorded along a path
including a turn.
The same accuracy was not achieved when using real data in real turning cases.
This can be seen in Figure 13 showing f , f , fˆ and fˆ when driving the
R F R F
routes seen in Figure 14, starting at (0,0) m with heading 0o. Test 3 is the only
case when both offsets are within the ±0.5o limit and close to the true offsets.
Test 2 is on the border of the ±0.5o range with f and close but outside with
F
f . In test 1 only f is in the range but by looking at fˆ in Figure 13, it’s seen
R R F
that
fˆ
after the turn, seen in Figure 14, converges towards the true offset. It
F
could also be seen that
fˆ
in test 3 deviates at the turn after 60 m and then
F
returns. In test 2 the offset is deviating all the time from the true but especially
when the turn is made. The computation time is around 15 minutes.
39 |
Chalmers University of Technology | 7.2.3 Driving in a straight wide mine drift
Simulations from Kvarntorp gave offsets (f ,f ) within ±0.2o of the true
R F
offsets. The simulations also showed that the effects of the small turns in the
end are neglect able.
Figure 15 shows f , f , fˆ and fˆ estimated from data recorded in
R F R F
Kvarntorp. For all the Kvarntorp runs the offsets are the same but not known
and it’s hard by just looking at Figure 15 to see what the true offsets could be
because of the large spread. The worst ones are the rear offsets which are
spread ±0.25o at all times. Even though it’s hard to see the true offsets it can be
seen in Figure 15 that (f ,f ) are inside the ±0.5o limit after 23 meters if the
R F
true offsets are somewhere between the estimated ones. The computation time
is around 20 minutes.
From looking at
(fˆ ,fˆ
) in Figure 15 it could be seen that for both offsets test
R F
1, 3 and 5 are grouped together and similar with test 2 and 4. Figure 16 shows
the driven paths super imposed on maps of the environment. All paths begin in
position (0,0) m with heading 0o. In tests 1, 3 and 5 the machine was driving
forward, while 2 and 4 are recorded while driving backwards. It’s thereby a
correlation between the direction and especially the rear offsets. From Figure
15 it can be seen that it takes 50 m for
(fˆ ,fˆ
) to stabilize and because the
R F
paths are not longer than 80 m (f ,f ) do not have time to stabilize at the final
R F
value. The noise on the maps in Figure 16 is because of holes in the wall into a
side drift.
Figure 16: The 5 maps and paths, in blue, from Kvarntorp. The paths begins at
position (0,0) m with heading 0o.
41 |
Chalmers University of Technology | 7.3 Discussion
In Chapter 6 it was found that an accuracy of ±0.5o was needed to get a scan
matching error less than 0.35. It’s here seen that it’s possible in most cases to
achieve this accuracy after only 23 m if the machine is going straight. Possibly
it could have been achieved faster if the mean valuing had started earlier since
the offsets
(fˆ ,fˆ
), according to Figure 11, in most cases have converged after
R F
only a few meters. The computation time is also less than the required 30
minutes.
The reason for tests 3 and 4 in Figure 11 to lie outside the allowed range can’t
be properly explained. Tests 1 and 8 are made in the same drift and in test 1
there are only minor problems to converge in the beginning. In test 8 there are
larger problems in the beginning and the start position is in the same place as in
tests 3 and 4. It may be that the machine is starting in a junction, resulting in the
estimated offsets getting stuck in local minima.
It could also be observed that the wall is noisy after half the path. This may also
be the reason why tests 3 and 4 don’t converge after the junction. Investigations
of e have shown that it decreases for tests 3 and 4 compared to in test 8 at
match
the place of the noisy wall, which strengthens the theory. This happens because
the noisy wall changes the optimal point searched for by the Kalman filter. The
noise may come from a cable ladder or ventilation drum. What is causing the
constant offsets between the tests in the same areas has yet to be proven.
Probably it has to do with differences in the starting point, steering of the
machine, noise in the gyro and so on.
A probable reason for
(fˆ ,fˆ
) to deviate from their true offsets in the
R F
simulated and real turns in Figure 12 and Figure 13 is because of geometry of
the turn, as mentioned in Chapter 6.3. The theory is strengthen by the fact that
fˆ
in one of the simulated cases isn’t affected until after the turn, and that we
R
don’t have simulated odometry errors. Test 2:s behaviour, seen in Figure 13,
can’t be explained. Something makes the offset deviate even on the straight
parts.
Later examination of the machine used at data collection in Kvarntorp showed
that a part in the hinge angle sensor was broken. This gave the hinge angle
sensor a play. The load on the hinge angle changes when the machine goes in
different directions, which then probably affected its readings because of the
play. This is probably the reason for the grouping in the tests seen in Figure 15.
Positioning of the rear part of the machine depends directly on the hinge angle,
which can be the reason for the rear offsets to get more affected.
The algorithm works in all simulated cases which mean that it is theoretically
correct. More experiments in a controlled environment are needed to draw some
definitive conclusions for the real cases but from the experiments made it can
be concluded that the offsets in most cases can be estimated to within ±0.5o, if
the environment doesn’t contain too much disturbance. The requirements are
42 |
Chalmers University of Technology | that the machine is driven straight for at least 23 m at a speed of around 2.5 m/s.
It even seems to work in wide drifts with side drifts. The computation time
increases with the width of the drift but stays below 30 minutes.
8 Conclusion
This report presents two methods that have been developed for use in
calibration of the hinge angle sensor respectively the two laser scanners on an
Atlas Copco ST14 LHD. In both the hinge angle calibration and laser scanner
calibration the new methods finds the angular offset of the sensors.
For hinge angle calibration we have developed an algorithm based on the
difference between the measured hinge angle value and the expected value
modelled from the speed and the heading change as measured by the gyro. In
the laser scanner calibration we used state augmentation together with an
already existing Kalman Filter to estimate the angular offsets of the laser. A
SLAM algorithm was also implemented to solve the problem with cross
dependence between the map and the offsets at the calibration.
Both methods satisfy the predefined criterions that no surrounding equipment
should be needed and no extra sensors should be needed. They also satisfy the
requirement of easy operation. The operator only needs to drive approximately
straight for around 50 meters while recording data and then supply the
calibration program with the log-file. Both methods can also operate offline and
deliver estimates in under 30 minutes as requested.
To test accuracy and dependence on outer parameters a number of experiments
and tests were made with both methods. The dependence on speed, distance and
steering angle was analytically investigated for the hinge angle calibration to
get the characteristics of the system. Real experiments were then made to
confirm the characteristics and find the limits of the system. For the laser
scanners the required precision was only known as a maximum scan matching
error in the map and not as a sensor offset. An investigation was therefore first
made to find the offsets that gave the maximum scan matching error. The
system was to complex to do analytical investigations, instead simulations and
real tests were made to test the accuracy of the system and how long driving
distance and time that was needed to do the estimation. The tests were made in
normally sized tunnels, tunnels with a turn and wide tunnels with side drifts to
test how the method handle different environments.
The experiments showed that the hinge angle calibration method is only able to
deliver accurate estimates two within ±0.3o in 33 m or more. It is also required
that the machine drive straight for this to be achievable. The tests showed that
the laser calibration method is able to deliver the required accuracy of ±0.5o in
around 15 minutes, and that a distance of 23 m in most cases is sufficient if the
machine is driven straight. Problems arise if there are to much environmental
disturbance or if the machine makes large turns. For both methods a speed of
2.5 m/s is recommended.
43 |
Chalmers University of Technology | The laser calibration method and the hinge angle method both need further
evaluation in the future. More experiments needs to be done to better establish
the confidence interval of both methods. As explained above side drifts
interfere the laser offsets estimation and maybe also the positioning so it needs
to be investigated further.
In Atlas Copco auto traming system maps are today created using only
odometry. In the future one could use the maps from the laser calibration
method instead where the positioning is made using both odometry and laser
measurements. One could also augment the hinge angle sensor offset to the
states and use the laser calibration method for the hinge angle offset estimation.
The hinge angle method could also be run online, when the offset is known to
keep track of the gyro and hinge angle sensor to see if an error occur, because
then the method deliver a different value compared to the true offset.
Finally the developed methods considerably simplify the calibration of the three
sensors compared to the methods used before. Both methods delivered such a
good results that Atlas Copco choose to implement them in their system.
9 References
Altafini, C. (1999). A path-tracking criterion for and LHD articulated vechicle.
International Journal of Robotics Research, Vol. 18, No. 5, pp. 435-441. May
1999.
AutoMine (2011). Available: http://www.miningandconstruction.sandvik.com.
[2011-02-02]
Borenstein, J. (1996). Measurements and Correction of Systematic Odometry
Errors in Mobile Robots. IEEE Transactions on robotics and automation. vol.
12, No. 6, pp. 869-880. December 1996.
Dall Larsen, T., Bak, M., A. Andersen, N. & Ravn O. (1998). Location
Estimation for and Autonomously Guided Vehicle using and Augmented
Kalman Filter to Autocalibrate the Odometry. IEEE International Conference
FUSION. (pp. 245-250). Las Vegas, Nevada, USA 6-9 July 1998.
De Ceccco, M. (2002). Self-Calibration of AGV Inertial-Odometric Navigation
Using Absolute-Reference Measurements. IEEE Instrumentation and
Measurement Technology Conference. (pp. 1513-1518). Anchorage, Alaska,
USA 21-23 May 2002.
Hainsworth, D.W (2001). Teleoperation User Interfaces for Mining Robotics.
Autonomous robots, vol. 11, ss. 19-28.
Larsson, J., Broxvall, M. & Saffiotti, A. (2010). An evaluation of local
autonomy applied to teleoperated vehicles in underground mines. IEEE
International Conference on Robotics and Automation (ICRA) (pp. 1745-
1752). Anchorage, Alaska, USA 3-8 May 2010.
Madhavan, R., Dissanayake, M.W.M.G. & Durrant-Whytel, H.F. (1998).
Autonomous Underground Navigation of an LHD using a Combined ICP-EKF
44 |
Chalmers University of Technology | Nitrogen removal in process water from the Gerum tunnel
A lab analysis using nitrification and denitrification as biological treatment
Master of Science Thesis in the Master’s Programme Infrastructure and
Environmental Engineering
VICTORIA LILJEDAHL
Department of Civil and Environmental Engineering
Division of Water Environment Technology
Chalmers University of Technology
ABSTRACT
High nitrogen concentrations in process water from tunnel constructions have become
an issue due to residues from explosive, transported to recipients in the construction
area. Biological treatment in terms of nitrification and denitrification are necessary to
remove the nitrogen compounds in the process water. Skanska Sweden AB is
currently constructing a tunnel in Tanumshede called Gerumstunneln, for which the
biological treatment of the process water has not worked during the project why this
study was initiated. The aim of this master thesis was to assess which factors and
parameters affect the biological treatment of process water in tunnel construction with
regard to nitrification and denitrification. Lab tests on process water from the tunnel
project were performed to achieve a complete nitrification and denitrification.
Three nitrification tests and one denitrification test were performed in the lab where
the first nitrification test used process water from two separate days and sludge that
was collected from a wastewater treatment plant. The nitrification process was
incomplete since nitrate did not increase even though there was a slight decrease in
ammonium and a slight increase in nitrite. The second test comprised of the same
water and sludge materials as Test 1 with the exception of adding sodium dihydrogen
phosphate, thereby increasing the phosphate concentrations which were thought to
limit the reactions. This test also failed to show any nitrifying activity. The alkalinity
measured in the process water was very low and for the third test, alkalinity and
phosphate were therefore increased. This test gave the best results of the nitrification
tests and resulted in ammonium being converted to nitrite, however the process failed
to transform nitrite to nitrate. The results from the denitrification test did show a
denitrifying activity when nitrate was transformed to nitrite at the same time as total
nitrogen (TN) decreased which indicates that nitrogen gas has been formed, though
the test should have been carried out for a longer period of time. Further lab analysis
on process water in tunnels with high nitrogen content is crucial which should focus
on metal concentration and the proportions of sludge and process water since those
parameters were not thoroughly analyzed in this study.
Key words: nitrogen, nitrification, denitrification, process water, tunnels, inhibiting
factors, water assessment
II |
Chalmers University of Technology | Kväverening av processvatten från Gerumstunneln
En labbanalys där nitrifikation och denitrifikation används som biologisk rening
Examensarbete inom Infrastructure and Environmental Engineering
VICTORIA LILJEDAHL
Institutionen för bygg- och miljöteknik
Avdelningen för Vatten Miljö Teknik
Chalmers tekniska högskola
SAMMANFATTNING
Höga kvävehalter i processvatten från tunnelkonstruktion har blivit ett problem på
grund av rester från sprängmedel som transporteras till vattenrecipienter inom
byggområdet. Biologisk behandling såsom nitrifikation och denitrifikation är
nödvändigt för att avlägsna kväveföreningar i processvattnet. Skanska Sverige AB
bygger för närvarande en tunnel i Tanumshede, Gerumstunneln, där den biologiska
reningen av processvattnet inte har fungerat under projektet varför denna studie
påbörjades. Syftet med detta examensarbete var att bedöma vilka faktorer och
parametrar som påverkar den biologiska reningen av processvatten under
tunnelkonstruktion med avseende på nitrifikation och denitrifikation.
Laboratorietester med processvatten från tunnelprojektet utfördes för att uppnå
fullständig nitrifikation och denitrifikation.
Tre nitrifikationstester och ett denitrifikationstest utfördes i labb där det första
nitrifikationstestet använde processvatten från två olika dagar och slam som samlats in
från ett avloppsreningsverk. Nitrifikationsprocessen var ofullständig då nitrat inte
ökade trots att det fanns en liten minskning av ammonium och en liten ökning av
nitrit. Det andra testet bestod av samma vatten - och slammaterial som Test 1 med
undantaget att natriumdivätefosfat tillsattes, vilket ökade fosfatkoncentrationerna so
ansågs begränsa reaktionerna. Detta test misslyckades också med att visa någon
nitrifierande aktivitet. Alkaliniteten som analyserades i processvattnet var mycket låg
och till det tredje testet ökades både alkalinitet och fosfat. Detta test gav de bästa
resultaten av nitrifikationstesterna och resulterade i att ammonium omvandlandes till
nitrit, dock så misslyckades testet med att omvandla nitrit till nitrat. Resultatet från
denitrifikationstestet visade en denitrifierande aktivitet när nitrat omvandlades till
nitrit samtidigt som totalkväve minskade, vilket indikerar att kvävgas har bildats, dock
borde testet ha utförts under en längre tidsperiod. Ytterligare labbanalys av
processvatten från tunnlar med hög kvävehalt är viktigt vilket borde fokusera på
metallkoncentrationen och proportionerna av slam och processvatten eftersom dessa
parametrar inte grundligt analyserats i denna studie.
Nyckelord: kväve, nitrifikation, denitrifikation, processvatten, tunnlar, inhibiterande
faktorer, vattenanalys
III |
Chalmers University of Technology | PREFACE
This master thesis has been conducted in the master’s program Infrastructure and
Environmental Engineering for Skanska Sweden AB - Infrastructure with the aim of
assessing nitrogen treatment during tunnel construction. The thesis was carried out
from January to May 2014 with guidance from both Skanska Sweden AB and the
department Water Environment Technology at Chalmers University of Technology.
I would like to thank my supervisor Britt-Marie Wilén at Chalmers University of
Technology for taking the time to help and guide me through the laboratory work. I
would like to acknowledge my supervisors at Skanska Sweden AB, Mikael Andersson
and Jenny Karlsson. Thank you for giving me the opportunity to conduct my thesis at
Skanska Sweden AB and for helping me to select this topic for my thesis. I would also
like to acknowledge John Nyberg at Skanska Sweden AB for your optimism,
guidance and interest in this topic, thank you. This thesis would not have been
completed without the help of Magnus Hallberg at Halfor AB, thank you for your
ebullient commitment, guidance and competence throughout this time. Further, I
would like to thank the “water boy” Niclas Johansson at Skanska Sweden AB in the
project E6 Pålen - Tanumshede. Thank you for all the help with water collection,
water samples and analysis at site. I would also like to acknowledge Mats Bäckman
and Charlotte Bougart at Hammargårdens treatment plant. A final thank to my
opponents Erik Eidmar and Johan Hultman for your valuable input and feedback on
my work.
This project has been very interesting and it has contributed to increase my knowledge
about the complexity of water contaminants and treatment processes which is a
worldwide issue. It is important to keep our waters clean, today and tomorrow, for us
and our next generations.
Göteborg, May 2014
Victoria Liljedahl
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
VI |
Chalmers University of Technology | 1 INTRODUCTION
Water is an essential part of life and only 2.5 percent of water resources around the
world are available freshwater. Sweden is one of few countries in the world where
freshwater availability currently is not a big problem. Sweden is covered by
approximately 10 percent of inland waters in terms of rivers, streams and lakes (SCB,
2012). There are 119 main watercourses with catchment areas bigger than 200 km2
that end up by the coast. There are in total 27,663 watercourses and the total length of
all watercourses is 192,000 km (SMHI, 2010). Hence, surface water is not a rarity in
the Swedish landscape. But even with sufficient groundwater and surface water, a
sustainable management of the freshwater resources is still very important. Especially
for future generations as the aim for Swedish waterways is to be ecologically
sustainable (Naturvårdsverket, 2014).
Due to this high areal coverage of surface water, the development of infrastructure
can be challenging and complex. Infrastructural construction in or near water in
Sweden is therefore inevitable. Tunnels are one construction that enables easy access
to and from different locations for commute users in the Swedish landscape, both in
terms of crossing waterways but also through rock passes. There is, for example, the
Tingstad tunnel in Gothenburg that crosses Göta Älv, the Northern Link in Stockholm
and the controversial railway tunnel through Hallandsås all of which are developed to
increase the infrastructure efficiency. Tunnel construction is increasing in Sweden and
there are currently around 20 road tunnels (Vägverket, 2001), which are mostly
concentrated to Stockholm, Gothenburg and the West coast and there are also 168
railway tunnels around the nation (Trafikverket, 2014).
There are many types of tunnel constructions in Sweden but the most common and
least expensive is a rock tunnel executed by blasting (Vägverket, 2001) where most
explosives used today contain ammonium nitrate, sodium nitrate and calcium nitrate
as the oxidizing agent (Forsyth, et al., 1995). One issue with the explosives is the
significant required concentrations of nitrogen (N) which is transferred or leached
through the runoff to recipients near the tunnel construction (Forsyth, et al., 1995).
The anthropological impact on the tunnel construction area, which leads to excessive
nitrogen concentrations in the runoff water, could if untreated have fatal impacts on
the aquatic life as well as declined quality of freshwater sources. With the large areal
coverage of waterways in Sweden, the risk of contaminating the water sources during
construction work is high and need cautious handling, such as efficient water
treatment at the construction site. The water treatment technologies in operation today
do not always meet the requirements of different parameters, especially treatment of
nitrogen compounds during colder temperatures.
1.1 BACKGROUND
Skanska Sweden AB is currently running a highway project in western parts of
Sweden called E6 Pålen-Tanusmhede. This project started in 2013 and aims to expand
and relocate the last phase of the highway E6 which runs from Pålen to Tanumshede.
The new stretch of E6 in that area requires a tunnel consisting of two tubes which is
performed by blasting. Blasting processes requires explosives which contain nitrogen,
and the contaminants in particular are ammonium and nitrate. Nitrogen has become an
issue due to its high concentrations in the water runoff from the tunnel construction
which mainly leads to water contamination in two ways. The first contamination
source is in the front of the tunnel where water is used to wash the blasted rock
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
1 |
Chalmers University of Technology | masses in order to settle dust and remove blasting residues. The other contamination
source is at the disposal site where crystalline nitrogen residues are kept which is
infiltrated to the groundwater or transported to the nearby recipients during
precipitation. The area has suffered from quite heavy precipitation which has
increased the concentration of nitrogen in the runoffs, thus putting higher pressure on
the water treatment facility. The biological treatment for nitrogen removal during the
project has not worked, which was presumed to be related to low water temperatures.
However, other inhibiting parameters could cause the nitrogen removal failure.
1.2 AIM
The aim of this master thesis is to assess which factors and parameters that affect the
biological treatment of process water in tunnel construction with regard to nitrification
and denitrification. The tunnel project Gerumstunneln in Tanumshede, Sweden, is
used as a case study due to arisen difficulties of meeting acceptable nitrogen
concentration in the recipients. The wastewater is a combination of process water
from the tunneling construction, stormwater runoff and leachate water, which
becomes contaminated by excessive nitrogen concentrations from explosive residues
during blasting and storage. One issue is the water temperature which is below 10°C
at which nitrification and denitrification processes are significantly decelerated, which
limits the nitrogen removal rate. Thus, the aim is to achieve a complete nitrification
and denitrification process in the lab by using the process water from the project.
1.3 SCOPE
This master thesis is carried out for Skanska Sweden AB due to heightened
restrictions of nitrogen concentration in process water discharge to recipients during
tunnel constructions. The thesis first presents background information regarding
nitrogen removal, conventional wastewater treatment and the case study. This is
followed by lab analysis descriptions and its results, which in turn is used for
calculations of which scale the water treatment facility need to have for different
temperatures. The thesis finalizes with a discussion and a conclusion which provides
necessary information to achieve sufficient nitrogen removal.
1.4 METHOD
This master thesis comprises of two parts. The first part consists of a literature study
with information about the current wastewater treatment method in Tanumshede and
other technologies that could potentially be of relevance for tunnel constructions. The
theory behind nitrogen is also presented in order to fully understand the different toxic
compounds. The second part is a lab analysis where water samples from Tanumshede
will be used to characterize the process water’s composition and conduct some
smaller experiments to assess if the process water can be nitrified and denitrified with
activated sludge from a nutrient removal wastewater treatment plant.
1.5 LIMITATIONS
The primary focus will be on nitrogen compounds even though other parameters, such
as heavy metals or phosphorus, could be factors that could inhibit the nitrification and
denitrification in process water from other tunnel constructions. The nitrogen
compounds in focus are ammonium nitrogen, nitrite nitrogen and nitrate nitrogen.
Throughout the report when discussing ammonium, nitrite and nitrate, they are always
referred to the forms of ammonium nitrogen , nitrite nitrogen and
𝑁𝐻4 −𝑁 𝑁𝑂2 −𝑁
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
2 |
Chalmers University of Technology | nitrate nitrogen , not to be mixed with , or . Thus, if
ammonium is mentioned, it is in the form of ammonium nitrogen, otherwise it will be
clearly stated. 𝑁𝑂3 −𝑁 𝑁𝐻4 𝑁𝑂2 𝑁𝑂3
The lab analysis will be conducted in a smaller scale and cannot fully represent the
actual treatment process in full-scale, which is more complex. The lab is also
performed during a limited time which could give different results compared to
studies under longer periods. The sludge needs to be collected outside of Gothenburg
which also limits the time and day for when the lab can be carried out. Due to this
geographical limitation, the sludge will be kept for a few days between some
experiments and will not always be used the same day as it was collected from the
treatment plant. There are geographical limitations for the process water as well since
the tunnel project is located approximately 135 km from Gothenburg. The process
water will most likely not contain much biological material why the same water will
be used throughout all experiments, but it will be stored in a cool place in between.
The focus in the lab will only be on nitrification and denitrification with some
alterations between the different tests in order to obtain different results. There are
many factors that could contribute to inhibition of nitrification and denitrification,
however, since the time period for this thesis is limited, only a few factors will be
analyzed. The water temperature could affect the nitrogen treatment and in reality, the
temperature is around 10°C in the tunnel water. However, the tests in the lab are
carried out in room temperature to observe if nitrification and denitrification could be
achieved at all and to assess other possible factors that could decline the nitrifying and
denitrifying activity.
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
3 |
Chalmers University of Technology | 2 NITROGEN THEORY
The largest source of nitrogen on the planet is the atmosphere which contains around
78 percent of nitrogen. Plants are dependent on nitrogen for growth and development
and the main source in soils are organic matter, which in turn originates from animal
and plant residues. However, plants need inorganic forms of nitrogen, thus bacteria in
the soils convert organic nitrogen to inorganic forms before being taken up by the
plant roots. Humans and animals eat the plants which are later returned to the soil in
residues (Killpack & Buchholz, 1993). There is a deficit of nitrogen in soil which
slows down the decomposition process of dead organic substances, which in turn
decrease the emissions of greenhouse gases (Golubyatnikov, et al., 2013). The
nitrogen cycle is presented in Figure 2.1. This shows that atmospheric nitrogen is
transported to the soils and plants. Animal residues decompose to the nitrogen
compound ammonium which is nitrified to nitrites and thereafter nitrates before
denitrified back to the atmosphere (U.S. EPA, 2012). The nitrification and
denitrification process are further described in Section 2.2.1 and 2.2.2. Rain
contributes atmospheric nitrogen when it reaches the soil and could reach the
groundwater aquifers. Nitrogen from factories producing commercial nitrogen
fertilizers could also enter the nitrogen cycle through farming and gardening.
Figure 2.1. The nitrogen cycle (U.S. EPA, 2012)
In tunnel construction it is important to monitor the discharge water from the
construction process. There are many parameters that should be carefully observed
such as total nitrogen, pH, alkalinity, sulphite, nitrate, nitrite, ammonium, suspended
particles and conductivity (Trafikverket & Ramböll, 2012). However, many
parameters are treated to designated limits and do often not constitute a problem. The
exception is nitrogen contaminations which is the primary nutrient of focus in this
master thesis due to high concentrations in the runoffs during tunnel construction.
Nitrogen is essential for growth of microorganisms, plants and animals and also in the
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
4 |
Chalmers University of Technology | synthesis of protein. The principle sources of nitrogen compounds in wastewater are
plant and animal origins, sodium nitrate and atmospheric nitrogen. Most nitrogen
compounds found in soil and groundwater are biologically originated
(Tchobanoglous, et al., 2004) but during tunnel construction, excessive nitrogen
compounds originate from the explosives.
2.1 NITROGEN COMPOUNDS
There are several nitrogen compounds of importance but the most common in general
are ammonia (NH ), nitrite (NO -) and nitrate (NO -) (Shrimali & Singh, 2001). The
3 2 3
most common compounds in wastewater are ammonia, ammonium, nitrogen gas,
nitrite ions and nitrate ions (Tchobanoglous, et al., 2004). Table 2.1 summarizes some
of the most common nitrogen compounds where the primary compounds in tunnel
construction are ammonium, nitrite and nitrate.
Table 2.1. Different forms of nitrogen and their definitions, modified from (Tchobanoglous, et al., 2004)
Forms of nitrogen Abbreviation Definition
Ammonia gas
Ammonium ion
𝑁𝐻3 𝑁𝐻3
+ +
Total ammonia nitrogen
𝑁TA𝐻N4 𝑁𝐻4
+
Nitrite
𝑁𝐻3 +𝑁𝐻4
− −
Nitrate
𝑁𝑂2 𝑁𝑂2
− −
Total inorganic nitrogen
T𝑁I𝑂N3 𝑁𝑂3
+ − −
Organic nitrogen Organic N 𝑁TK𝐻N3 +– (𝑁𝐻4 +𝑁𝑂2 )+ 𝑁𝑂3
+
Total Kjeldahl nitrogen TKN Organic
𝑁N𝐻3+𝑁𝐻4
+
Total nitrogen TN Organic
N+𝑁𝐻3+𝑁𝐻4
+ − −
Nitrogen removal is correlated to temperature where +th𝑁e𝐻 r3 e+ac𝑁ti𝐻o4 n +ra𝑁te𝑂s2 in+c𝑁re𝑂a3 se with
temperature. Chang, et al., (2013) showed how the nitrogen removal efficiency
changed during the four seasons for nitrate ( ) and total nitrogen (TN) which
turned out to be as expected, the highest removal −efficiency during summer and the
lowest during winter using the IVCW
(inte𝑁gr𝑂a3
ted vertical constructed wetland)
system. Vertical constructed wetlands can have up-flow or down-flow and IVCW
combines these two wetlands to improve the water quality. Wetlands consist of plants
and some sort of media, e.g. gravel, where the wastewater runs through the media
(wetland) (Peng, et al., 2014).
It is important to separate ammonium with ammonium nitrogen for example in terms
of concentrations. It is the same nitrogen compound, but when concentrations are
analyzed, it differs between the two types. If a concentration is presented as
ammonium, a factor is required to convert this concentration to ammonium nitrogen.
The atomic mass for nitrogen is 14, hydrogen 1 and oxygen 16. This is used to
calculate the factors for each nitrogen compound. These factors are presented in
Equations 2.1-2.4.
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
5 |
Chalmers University of Technology | (Eq 2.1)
14 14
𝑁𝐻3 → 𝑁𝐻3 −𝑁:1×14+(3×1) = 17 = 0.82 (Eq 2.2)
14 14
𝑁𝐻4 → 𝑁𝐻4 −𝑁:1×14+(4×1) = 18 = 0.78 (Eq 2.3)
14 14
𝑁𝑂3 → 𝑁𝑂3 −𝑁:1×14+(3×16) = 62 = 0.23 (Eq 2.4)
14 14
𝑁𝑂2 → 𝑁2𝑂.12.−1 𝑁:N1I×TR14I+T(2E× 1 6) = 46 = 0.3
Nitrite is considered to be quite unstable and is easily oxidized to nitrate. Nitrite
is measure−d colorimetrically and it is an indicator of past pollution. Nitrite is often
found i𝑁n 𝑂sm2 all concentrations, rarely above 1 mg/l in wastewater, but it is none the
less very toxic to fish and other aquatic animals. If nitrites are found in wastewater,
they are oxidized with chlorine in wastewater treatment plants in order to be reduced
(Tchobanoglous, et al., 2004). If nitrite is present in high concentrations in water and
if the water is ingested by human or animals, then nitrite can turn haemoglobin to
methemoglobin which can cause anoxia and death since methemoglobin cannot
transport oxygen to cells (Alonso & Camargo, 2003), also known as the blue baby
syndrome.
2.1.2 NITRATE
The most oxidized form of nitrogen in wastewater is nitrate of which the
concentrations could be measured using specific-ion electrodes, o−r as nitrite with
colorimetric methods. The U.S. EPA limits nitrate in drinking
w𝑁a𝑂te3
r to 45 mg/l and
the wastewater effluent concentration often varies between 0 and 20 mg/l. High
concentrations of nitrates in water can have serious and even fatal effects for infants
(Tchobanoglous, et al., 2004).
Camargo et al. (2005) shows that nitrate can be removed by the use of for instance
aquatic plants, algae and bacteria but also by using ion exchange, reverse osmosis
(RO) and electro-dialysis since the conventional processes such as chlorination, UV
and filtration are not viable for nitrate ions (Shrimali & Singh, 2001). However, the
treatment processes are quite expensive and some of them rather cause new problems
in terms of nitrate disposal.
2.1.3 AMMONIA AND AMMONIUM
Ammonia gas is the oxidation state in most organic compounds. Ammonia can
be present as either ammonia gas or ammonium ion in aqueous solutions,
which depends𝑁 𝐻o3 n the pH. The concentrations of ammo+nia can be measured
colometrically, titrimetrically or w𝑁it𝐻h 3 specific-ion electrod𝑁es𝐻 4 (Tchobanoglous, et al.,
2004). The un-ionized molecular ammonia exists in equilibrium with the ionized
ammonium dissolved in water which depends on both pH and temperature. The
existence of ammonia in environments with a pH below 7 is very low, hence the
ammonia ions increases with increased pH (U.S. EPA, 1993). If the concentration of
ammonia in waterways is in the range of 0.2-0.5 mg/l, the toxicity to fishes can be
fatal (Miladinovic & Weatherley, 2008).
2.1.4 ADVERSE EFFECTS FROM NITROGEN COMPOUNDS
Three major environmental problems have been identified by Camargo & Alonso
(2006) of which all include inorganic nitrogen. Eutrophication is one problem,
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
6 |
Chalmers University of Technology | acidification of ecological ecosystem the second problem and the third is direct
toxicity which affects aquatic life’s survival, growth and reproduction.
EUTROPHICATION
Increased concentrations of various nitrogen forms in freshwater could generate in
eutrophic conditions. Eutrophication presents itself with excessive plant growth and
algae blooming which is the result of over-fertilization in water bodies. With
increased plant growth, reduced DO and non-clear waters, the organisms in the water
body receives less sunlight and less oxygen since it is being depleted by the growing
plants and algae, hence the survival of the aquatic life could be critical (U.S. EPA,
1993). If wastewater from tunneling shows increased nitrogen values, the recipient
could potentially indicate increased nutrient growth, hence degrading the water
quality. Eutrophication in water resources is a worldwide problem in both freshwater
and marine environment. In Sweden, the problem is concentrated in the southern parts
but some mountain areas indicate eutrophication as well (Naturvårdsverket, 2013).
There are many adverse effects on freshwater, some includes increased productivity
and biomass of phytoplankton and suspended algae, shifts in phytoplankton
composition to toxic species, threats to aquatic species, problems in drinking water
supply, depletion of oxygen and decreased recreational values (Smith, 2003). It is not
only the aquatic life and environmental surroundings that are affected, humans could
also be threatened. When blooming algae dies, algae’s toxin can be produced which is
harmful to humans. Inorganic nitrogen (ammonia, ammonium, nitrite and nitrate), in
particular nitrate, is used in the eutrophication reaction but total nitrogen (TN) is also
considered when assessing eutrophication. Nitrite concentration in eutrophic water is
also a threat to humans since the product of nitrite nitrification is carcinogenic (Yang,
et al., 2008). Nitrate could especially reduce eutrophication if it is decreased (Deng, et
al., 2012).
ACIDIFICATION
Nitrogen dioxide and nitrogen oxide are the main nitrogen acidifying
pollutants (sulphur dioxide is also a major pollutant) in lakes and streams. These
pollutants emit int𝑁o 𝑂th2 e atmosphere and can𝑁 t𝑂ransform by complex reactions into
sulfuric acid and
n𝑆i𝑂tr2
ic acid (Baker, et al., 1991). Anthropogenic
acidification of streams and lakes could cause many adverse effects e.g. biotic
impoverishmen𝐻t2 o𝑆f𝑂 4 salmonids and inve𝐻rt𝑁eb𝑂ra3 tes. Other adverse affects in freshwater
are increased accumulation and toxicity of aluminum (Sprenger & McIntosh, 1989),
hatching delay of fish and amphibian eggs, increased migration of aquatic insect from
their nests, declined zooplankton diversity and reduced growth rates in fish (Camargo
& Alonso, 2006).
TOXICITY
Ammonia concentrations of 0.1-1.0 mg/l are often seen as lethal for the wildlife.
Biological membranes are permeable to unionized ammonia which makes the toxicity
of ammonia a function of pH (Abeliovich, 1992). Ammonia is very toxic to the
aquatic wildlife, compared to the non- or less toxic ammonium (Camargo & Alonso,
2006).
Nitrite concentrations on the other hand should not exceed 0.1 mg/l in closed water
systems. However, lethal concentrations are much higher, around 250-350 mg/l
proved to be lethal to many fish, hence nitrite is less toxic than ammonia but should
still be kept under observations (Abeliovich, 1992). Though, other studies show that
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
7 |
Chalmers University of Technology | nitrite concentrations above 45 mg/l could result in anemia in infants and pregnant
women (Odjadjare & Okoh, 2010). The main toxic effect of nitrite on fish is hypoxia
which later results in death. This is due to a conversion of pigments former carrying
oxygen shifting to a state where they are incapable o carrying oxygen. Other effects of
nitrite on fish are cause of electrolyte imbalance, effects on membrane potentials,
skeletal muscle contractions and heart function, tissue shortage and repression of
the immune system (Jensen, 2003).
𝑂2
Nitrate toxicity has been considered irrelevant since it must be converted to nitrite
before it becomes toxic. The nitrate uptake is much more limited than nitrite uptake
which contributes to low nitrate toxicity. However, studies have shown that small
concentrations of nitrate can have adverse effects on sensitive aquatic animals during
long-term exposure. Further, freshwater animals tend to be more sensitive to nitrate
toxicity compared to marine animals (Camargo, et al., 2005).
When assessing the effects of increased nitrogen concentration in drinking water of
humans, methemoglobin could be the result from conversion of nitrates to nitrites
under anaerobic conditions in the digestive tract (Greer & Shannon, 2005). Symptoms
are headache, fatigue, stupor, convulsions and could lead to death. Ingested nitrates
and nitrites could be contributing factor of developing cancer of the digestive tract
(Fewtrell, 2004). Adverse affects from blooming toxic cyanobacteria on humans have
been reported around the world, including Sweden, with effects such as eye irritation,
fever, diarrhea, skin rash and muscular cramps. Some cases of human exposure led to
lethal outcomes where children are the most sensitive group ( (Hitzfeld, et al., 2000)
and (Chorus, 2001)).
As mentioned earlier, algae can cause toxicity to aquatic animals and these toxins can
either remain in the algal cells or be released into the water. Thus, animals can be
directly exposed to toxins through absorption, drinking or ingesting by feeding
activities. Cyanobacteria, diatoms and dinoflagellates are the predominate groups that
contribute to toxic algae (Camargo & Alonso, 2006).
2.2 BIOLOGICAL TREATMENT FOR NITROGEN
Water in tunnel constructions differ in composition compared to regular household
wastewater (e.g. the lack of faecal contaminations). A conventional wastewater
treatment plant consists of mechanical, chemical and biological treatment and
sometimes disinfection as a final step. Conventional wastewater treatment is
necessary as a pre-treatment step for nitrogen removal. The biological treatment step
use microorganisms such as bacteria in order to feed on the organic left-over matter
from the mechanical treatment step (The World Bank, n.d). Microorganisms feed on
nitrogen and around 20 percent is removed by flocculation and separation
(Naturvårdsverket, 2009). However, since all nitrogen is not fully removed in
biological treatment, a specific nitrogen removal step is added which is used in bigger
wastewater treatment plants serving more than 10,000 people and sensitive recipients
(Naturvårdsverket, 2009). Ammonia is converted into nitrate by nitrifying bacteria in
an aerobic condition after which denitrifying bacteria convert nitrate into nitrogen gas
in anoxic conditions (Gryaab, n.d). These nitrifying and denitrifying bacteria can be
obtained from sludge in a wastewater treatment plant. The anticipated nitrogen
removal efficiency is around 50 to 75 percent (Naturvårdsverket, 2009).
Nitrification and denitrification processes are the basis of all biological nitrogen
removal methods at the moment used for wastewater treatment. This has proved to be
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
8 |
Chalmers University of Technology | both economically and technically feasible for centralized and decentralized systems.
Aerobic processes (oxygen rich) nitrify ammonium, i.e. transform ammonium into
nitrate where after anoxic processes (oxygen absent) reduce nitrate to nitrogen gas
(Oakley, et al., 2010). Ammonium and organic nitrogen are the most common
compounds of nitrogen in a modern wastewater treatment plants and is removed by
biological nitrification/denitrification (Van Hulle, et al., 2010). Figure 2.2 presents
three different nitrification and denitrification processes where a) is a pre-anoxic
process starting with denitrification in an anoxic tank with wastewater as the carbon
source and is followed by nitrification in the aeration tank. The nitrified effluent is
Figure 2.2. Nitrification and denitrification processes, retrieved from
Oakley
later recycled back to the anoxic tank. In process b), which is a post-anoxic process,
the order is reversed from the pre-anoxic system and the carbon source can be either
external sources such as methanol or endogenous respiration of bacterial cells. In c)
the nitrification and denitrification occurs simultaneously in the same reactor (Oakley,
et al., 2010).
2.2.1 NITRIFICATION
The nitrification process is presented in Reaction 2.1 and 2.2 which describe the two-
step biological process of ammonia oxidized to nitrite followed by oxidation of nitrite
into nitrate (Tchobanoglous, et al., 2004). Reaction 2.1 shows how ammonia (or
ammonium) is converted into nitrite with ammonium-oxidizing bacteria (AOB) such
as Nitrosomonas as the most common group of bacteria while Reaction 2.2 converts
nitrite into nitrate by nitrite-oxidizing bacteria (NOB) such as Nitrobacter (U.S. EPA,
2002).
(Re 2.1)
− + −
(Re 2.2)
𝑁𝐻3 +𝑂2 → 𝑁𝑂2 +3𝐻 +2𝑒
− − + −
Reactions using the AOB-bacteria and NOB-bacteria can also be presented as
𝑁𝑂2 +𝐻2𝑂 → 𝑁𝑂3 +2𝐻 +2𝑒
Reaction 2.3 and 2.4 in the nitrification process which nitrifies ammonium to nitrite
and nitrate (Tchobanoglous, et al., 2004).
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
9 |
Chalmers University of Technology | (Re 2.3)
+ − +
(Re 2.4)
2𝑁𝐻4 +3𝑂2 → 2𝑁𝑂2 +4𝐻 +2𝐻2𝑂
− −
The total oxidation reaction is presented in Reaction 2.5 which shows how ammonium
2𝑁𝑂2 +𝑂2 → 2𝑁𝑂3
is oxidized into nitrate (Tchobanoglous, et al., 2004).
(Re 2.5)
+ − +
The nitrification process is not able to remove the toxic nitrogen compound ammonia,
𝑁𝐻4 +2𝑂2 → 𝑁𝑂3 +2𝐻 +𝐻2𝑂
but the process converts ammonia into nitrate as stated in the reactions presented
earlier. This reduces or eliminates the toxicity to fish as well as reducing the nitrogen
oxygen demand (NOD) of the effluent. The rate of the nitrification process in
wastewater is a function of time, and also, it is independent of the ammonia nitrogen
concentration (Viessman, et al., 2009).
A low growth rate characterizes nitrifying bacteria which is due to the low energy
yield, which in turn is connected to the oxidation of ammonia and nitrite. Most
nitrifying bacteria are autotrophic and use carbon dioxide as the carbon source, which
should be reduced before the carbon reacts with the cell mass. The reduction is
performed through oxidation of ammonium and nitrite hence the reaction for growth
is shown in Reaction 2.6 and 2.7 (Henze, et al., 1997).
(Re 2.6)
+ − +
(Re 2.7)
15𝐶𝑂2+13𝑁𝐻4 → 10𝑁𝑂2 +3𝐶5𝐻7𝑁𝑂2 +23𝐻 +4𝐻2𝑂
+ − − +
Equation 2.5 is a suggested expression by the U.S. EPA (1993) of the maximum
5𝐶𝑂2 +𝑁𝐻4 +10𝑁𝑂2 +2𝐻2𝑂 → 10𝑁𝑂3 +𝐶5𝐻7𝑁𝑂2 +𝐻
growth rate of Nitrosomonas as a function of temperature over the range 5-30°C.
(Eq 2.5)
0.098(𝑇−15)
𝜇𝑁 = 0 .47𝑒
−1
𝜇𝑁 = 𝑚𝑎𝑥𝑖𝑚𝑢𝑚 𝑠𝑝𝑒𝑐𝑖𝑓𝑖𝑐 𝑔𝑟𝑜𝑤𝑡ℎ 𝑟𝑎𝑡𝑒 𝑜𝑓 𝑁𝑖𝑡𝑟𝑜𝑠𝑜𝑚𝑜𝑛𝑎𝑠 𝑑
𝑒 = 𝑏𝑎𝑠𝑒 𝑜𝑓 𝑁𝑎𝑝𝑖𝑒𝑟𝑖𝑎𝑛 𝑙𝑜𝑔𝑎𝑟𝑖𝑡ℎ𝑚𝑠,2.718
Since the temperature is affecting the nitrification, winter aeration need to be much
𝑇 = 𝑡𝑒𝑚𝑝𝑒𝑟𝑎𝑡𝑢𝑟𝑒,°𝐶
longer than during summer. There are some ways to solve those seasonal issues, the
mixed-liquor suspended solids (MLSS) could be increased and pH could be adjusted
(Viessman, et al., 2009).
2.2.2 DENITRIFICATION
The denitrification process reduces the oxidized forms of nitrogen to nitrogen gas by
heterotrophic bacteria and by using a carbon source of biodegradable organic matter
(Mees, et al., 2014). First, there is a biological reduction of nitrate to nitric oxide NO,
then nitrous oxide and finally nitrogen gas N (Tchobanoglous, et al., 2004). The end
2
product, nitrogen gas, has no significant effect on the environment (U.S. EPA, 1993).
Denitrification occurs when oxygen is depleted, hence the process only occurs in
anaerobic or anoxic environments with enough quantities of nitrate and where the
oxygen demand surpasses the oxygen supply (Deng, et al., 2012). The nitrification
process has limited options for types of bacteria whilst denitrification has a broad
range of useful bacteria, which can be both heterotrophic and autotrophic. Some
heterotrophic organism genera are Achromobacter, Bacillus, Flavobacterium and
Psedumonas, where the latter is the most common denitrifier. Most of the
heterotrophic bacteria have the ability to use oxygen, nitrite or nitrate
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
10 |
Chalmers University of Technology | (Tchobanoglous, et al., 2004) though nitrite or nitrate is the wanted forms for the
denitrification process. However, if oxygen is present it will be used naturally instead
of nitrate and vice versa in anoxic environments. Nitrite and nitrate are considered to
be the electron acceptors in the respiratory electron chain, which is the primary
mechanism for energy generation in cells. Consequently, nitrite and nitrate replaces
oxygen in the transport chain which generates to some extent less energy (U.S. EPA,
1993). A simplified denitrification reaction is seen in Reaction 2.8 (Henze, et al.,
2002).
(Re 2.8)
− −
The denitrification process needs an organic carbon source to act as a hydrogen donor
𝑁𝑂3 → 𝑁𝑂2 → 𝑁𝑂 → 𝑁2
and to supply the biological synthesis. Some organic carbon sources could be acetic
acid, ethanol, methanol or organic matter. Methanol is preferred common source since
it is the least expensive synthetic compound that can be used due to its quality of not
leaving residual BOD in the effluent. Methanol first reduces DO in order for the
biological reduction of nitrate and nitrite to take place (Viessman, et al., 2009). These
different reactions, using methanol ( ) are presented in Reactions 2.9-2.11.
(Re 2.9)
𝐶𝐻3𝑂𝐻
(Re 2.10)
3𝑂2 +2𝐶𝐻3𝑂𝐻 = 2𝐶𝑂2 +4𝐻2𝑂
− −
(Re 2.11)
6𝑁𝑂3 +5𝐶𝐻3𝑂𝐻 = 3𝑁2 +5𝐶𝑂2 +7𝐻2𝑂+6𝑂𝐻
− −
The energy yielding process for denitrifying bacteria in Reaction 2.12 is a
2𝑁𝑂2 +𝐶𝐻3𝑂𝐻 = 𝑁2 +𝐶𝑂2 +𝐻2𝑂+2𝑂𝐻
combination of two half-expressions. The process uses organic matter in wastewater
as the energy and carbon source (Henze, et al., 1997).
(Re 2.12)
1 1 3 1 +
70𝐶18𝐻19𝑂9𝑁+5𝑁𝑂3 +5𝐻 →
1 17 1 − 1 + 1
𝑁2 +2.2𝐶.𝑂32 +TRE𝐻A𝐶T𝑂M3 E+NT 𝑁T𝐻IM4 +E SP𝐻A2𝑂NS
10 70 70 70 5
The trial spans are quite long for many of the researches in the literature with
examples of observation and analyze time of 180, 225 and 550 days for nitrification
and denitrification processes (Hoilijoki et al. 2000, Jokela et al. 2002 and Koren et al.
2000). Isaka et al. (2007) had a start-up period of 3 months before the nitrification
process was fully working whilst Jokela et al. 2002 had less than 3 weeks of start up
for nitrification and around 2 weeks for denitrification, though the entire analysis of
nitrification and denitrification proceeded over more than 225 days. Thus the time for
nitrification and denitrification can vary significantly.
2.2.4 NITRIFICATION AND DENITRIFICATION RATES
There are different nitrification and denitrification rates presented in several studies,
summarized in this section and in Table 2.2 and 2.3. One research using cold mine
water had nitrification rates of 29 and 38.3 g N/m3 at 5°C and 10°C respectively. The
nitrification rate at 5°C increased to 43 g N/m3 if the salinity was increased. The
denitrification rates were 605 and 522 g N/m3 for 5°C and 10°C (Karkman, et al.,
2011). Zaitsev et al. (2008) had a nitrification removal rate of 0.54 g NH -N/m2-d as
4
the highest at 5°C. The denitrification rate was 4.1 g NO₃-N/m2,d at 5°C. There were
also presented other denitrification rates at 3°C and 7°C of 1.5 g NO₃-N/m2-d and 1.6
g NO₃-N/m2-d respectively.
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
11 |
Chalmers University of Technology | One study performed by Rostron et al. (2001) achieved different nitrification rates for
CSTRs (continuously stirred tank reactors) which varied from 2.28 to 4.24 g N/m2-
d,media with the synthetic ammonia feed with a concentration of ammonia nitrogen
500 mg/l. There are some other nitrification rates found in the literature using
immobilized biomass. With the use of domestic sewage for example, a rate of 0.29 kg
N/m3-d was obtained with the ammonia nitrogen concentration of 50 mg/l. Several
studies have used synthetic feed water where the rates have been 0.58, 0.2, 0.25, 0.73
and 1.5 kg N/m3-d with respectively 27, 35, 50, 200 and 50 mg/l of ammonia nitrogen
(Rostron, et al., 2001). Hoilijoki et al. (2000) reported maximum nitrification rates of
0.13, 0.27±0.06 and 1.78 mg NO -N g/VSS,h.at 7, 10 and 24°C in December, August
3
and July respectively. This study was conducted using an aerobically pretreated
leachate.
Some in-situ nitrification rates during cold seasons (average of 10°C) varied for
different sites from 204 to 548 mg N/m2,d while the denitrification rates in the same
season varied from 180 to 491 mg N/m2,d. These rates are based on influent values of
62 mg/l for TN and 60 mg/l for NH -N. Sewage with low median strength was used in
4
the study for colder temperatures (Zimmo, et al., 2004). The nitrification rate in
sediment in lakes have been assessed which showed to be much higher than in the
water columns. The ammonia nitrogen concentration in the sample was 2-10 mg/l and
the nitrification rates in the sediments of two different lakes resulted in 0.32 and 0.37
g N/m2-d (Pauer & Auer, 2000).
The nitrification rate was analyzed in leachate from a landfill in Japan which had the
characteristics of 400-610 mg/l NH -N and 420-650 mg/l TN in the influent water.
4
The effluent water had NH -N concentrations of 16-35 mg/l which is a large decrease.
4
The highest nitrification rate was 0.71 kg N/m3-d at 10°C which took up to 3 months
to achieve (Isaka, et al., 2007) while Jokela et al. (2002) achieved nitrification rates of
0.05 kg N/m3-d at 5 and 10°C.
Table 2.2. Nitrification rates.1) (Kar kman, et al., 2011), 2) (Rostron, et al., 2001), 3) (Isaka, et al., 2007),
4) (Jokela, et al., 2002), 5) (Zimmo, et al., 2004) and 6) (Pauer & Auer, 2000).
Nitrification
Temp Rate Unit Type of Temp Rate Unit Type of
°C
°C water water
5 291 g N/m³ Mine water 5/10 504 g N/m³-d Leachate water
5 381 g N/m³ Mine water - 2902 g N/m³-d Syn. wastewater
10 431 g N/m³ Mine water 10 0,2045 g N/m²-d Wastewater
- 5802 g N/m³,d Syn. wastewater 10 0,5485 g N/m²-d Wastewater
- 2002 g N/m³,d Syn. wastewater - 0,326 g N/m²-d Lake sediment
- 2502 g N/m³,d Syn. wastewater - 0,376 g N/m²-d Lake sediment
- 7302 g N/m³,d Syn. wastewater - 2,282 g N/m²-d,med Syn. wastewater
- 15002 g N/m³,d Syn. wastewater - 4,242 g N/m²-d,med Syn. wastewater
10 7103 g N/m³,d Leachate water
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
12 |
Chalmers University of Technology | Table 2.3. Denitrification rates. 1) (Karkman, et al., 2011), 2)
(Zimmo, et al., 2004) and 3) (Zaitsev, et al., 2008)
Denitrification
Temp °C Rate Unit Type of water
5 5221 g N/m³ Mine water
10 6051 g N/m³ Mine water
10 0,182 g N/m²,d Wastewater
10 0,4912 g N/m²,d Wastewater
3 1,53 g NO₃-N/m²,d Mine water
5 4,13 g NO₃-N/m²,d Mine water
7 1,63 g NO₃-N/m²,d Mine water
2.2.5 SLUDGE FOR BIOLOGICAL TREATMENT
The sludge concentration varies throughout the literatures in different nitrification and
denitrification studies. The study performed by Koren et al. (2000) used 750 ml liquid
from activated sludge reactors which was mixed with 16 liters of water, thus around
4.5 percent of the total sample. The aeration flow was 2.4 l/min and the agitation rate
was 1 000 rpm. Rostron et al (2001) used nitrifying biomass of 2.5 g VSS/l using
CTSR (continuously stirred tank reactors) of 4 liters. Both Jokela et al. (2002) and
Hoilijoki et al. (2000) used nitrifying activated sludge of 4 g VSS/l when studying
leachate. Another study on wastewater in Hong Kong used a sludge concentration of 2
g MLVSS/l (mixed-liquid volatile suspended solids) in combination with aeration of 8
l/min (Li, et al., 2013).
2.2.6 INHIBITING FACTORS
Nitrification and denitrification can be inhibited for many reasons. Both occurrence
and rate of the nitrification process are controlled by environmental and operating
conditions, e.g. temperature and sludge age (Viessman, et al., 2009) and other
parameters such as phosphate concentration and pH have also been proved to limit the
nitrification and denitrification processes. Some of these factors will be considered
during the lab analysis.
TEMPERATURE
Nitrification and denitrification processes are strongly dependent on the temperature
where the rates of nitrification could double with a temperature increase of 10°C
while the denitrification process could double for every increase of 4°C (Zaitsev, et
al., 2008). Koren et al. (2000) states that nitrification is affected in temperatures below
10°C since the bacterial metabolism decreases and at temperatures below 4°C, the
bacterial metabolism stops all together. However, nitrogen removal is achievable at
temperatures below 10°C. The reduction rate is though decreased compared to room
temperature of 20°C. One study showed a 30 percent nitrification reduction at 10°C
compared to that of 20°C (Alawi, et al., 2009). Jokela et al. (2002), Hoilijoki et al.
(2000) and Zaitsev et al. (2007) showed that a nitrification efficiency of higher than
95 percent could be achieved with different loadings. However, other literature found
a 100 percent reduction in nitrification at 5°C and pointed out that high nitrification
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
13 |
Chalmers University of Technology | efficiency in low temperatures can only be achieved with long retention times or a
combination of increased nitrifier concentration and elevated temperature if it is
possible to heat the water (Ducey, et al., 2010).
The bacteria genera Nitrospira is favoured during cooler temperatures and increased
dissolved oxygen (Wells, et al., 2009). Adapting the nitrification process to lower
temperatures has been an issue in colder northern regions such as Sweden. Some
techniques have been tried and proved to be successful at 14°C and even at 7°C. They
include extension of the aeration period and supplemental additions of nitrifying
biomass in an aeration tank (Ward, et al., 2011). The nitrification rate decreases with
about 50 percent for every 10-12°C drop in temperatures above 10°C and when the
wastewater is even colder, e.g. decreasing the temperature to 5°C from 10°C, the
ammonia oxidation is halved (Viessman, et al., 2009).
NITROGEN COMPOUNDS
Nitrite-nitrogen can inhibit the biological process if the concentration is too high.
Nitrite-nitrogen concentration of 150 mg/l proved to inhibit the denitrification process
while 1 350 mg/l of nitrate-nitrogen inhibited the denitrification process. However,
inhibition of denitrification is not only dependant on nitrite concentration but the
biomass concentration. In an experiment using the same nitrite-nitrogen concentration
of 20-25 mg/l and a pH of 7-8, the denitrification was only completed with a biomass
concentration of 500-1 000 mg/l whilst using a biomass concentration of 100-150
mg/l, the denitrification was inhibited. It was also concluded in this study that a
decreased pH generates increased inhibition (Glass, et al., 1997). A critical nitrite-
nitrogen dose for nitrification is 50-100 mg/l at a pH 7 which caused inhibition.
Moreover, the inhibition lasted more than 10 days after the nitrite disappeared from
the solution (Philips & Verstraete, 2000).
Ammonia, i.e. molecular free ammonia, is the major compound responsible for
toxicity effects on aquatic life (U.S. EPA, 1993) and it can suppress the nitrifying
process. Toxicity can have great impacts on nitrifiers and can kill the nitrifying
bacteria. Free ammonia can halt the nitrification process in terms of being inhibiting
on the bacteria Nitrosomonas at free ammonia concentrations of 10-150 mg/l and at
0.1-1 mg/l for Nitrobacter (U.S. EPA, 1993). Other studies show that concentrations
of free ammonia higher than 0.2-1 mg/l inhibit the nitrite oxidation (He, et al., 2012).
PHOSPHATE
Phosphorus is most often present in the environment as organic or inorganic
phosphate. Phosphate has been reported to limit nitrification, especially during low
temperatures. In some nitrifying systems using filters, the phosphate concentrations
were dosed 5-50 µg (de Vet, et al., 2012). de Vet et al. (2012) performed a
study using filters. The3 −two different tests using filters with 80 µg had
complete nitrification𝑃 𝑂pr4 oce−ss𝑃es whilst the two other tests with 20 and 30 µg
resulted in incomplete nitrification. However, with concentration around𝑃 𝑂104 0− µ𝑃g, the
nitrification was inhibited as well.
𝑃𝑂4−𝑃
One pilot study by Kors et al. (1998) showed that by dosing with phosphate, the
ammonium removal increased in the nitrification process. With phosphate
concentrations below 15 µg /l in combination with very low temperatures, the
nitrification process failed to convert ammonium to nitrate. It was then assumed that
the phosphate concentration w𝑃a𝑂s 4 a limiting factor for ammonium removal besides the
temperature. When the influents in that study were dosed with 100-500 µg /l,
𝑃𝑂4
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
14 |
Chalmers University of Technology | ammonium removal rates increased despite temperatures around 1°C (Kors, et al.,
1998). Another pilot study also proved that phosphate was limiting the nitrification
process. When dosing10 µg , the ammonium removal rate increased and the
nitrification process could be com3−pleted (van der Aa, et al., 2002).
𝑃𝑂4 𝑃
METALS
Metals could also be inhibiting for the nitrification process. Some metals that have
been identified as inhibitors are zinc, copper, chromium, nickel and cobalt (U.S. EPA,
1993). Some inhibiting concentration ranges for certain metals have been identified
such as 0.08-0.5 mg/l for zinc and 0.005-0.5 mg/l for cupper, thus cupper inhibits the
nitrite and nitrate production greater than zinc (Juliastuti, et al., 2003). Juliastuti et al.
(2003) showed that when the concentrations of heavy metals increased, the inhibition
also increased on nitrification. For example, they concluded that a 97 percent
inhibition was reached when zinc had a concentration of 1.2 mg/l. Tchobanoglous et
al. (2004) also presented some concentrations at which the metals could be inhibiting;
0.25 mg/l nickel, 0.25 mg/l chromium and 0.10 mg/l copper have for example proved
to completely inhibit the ammonia oxidation. When it comes to nickel and cadmium, a
study showed that both nickel and cadmium hade higher inhibiting effects on the
ammonium oxidation rate than on the nitrite oxidation rate (Hu, et al., 2002).
pH
The pH can inhibit the nitrification rate. Nitrification should be performed in
conditions when the pH is around 6.5-8 but for optimal nitrification rates the pH
should be 7.5 to 8.0 according to the U.S. EPA (1993). The percent of the maximum
nitrification rates increases as pH increases from 5.0 to 8.0 before the maximum rates
starts to decline as the pH continues to increase from 8.0. With pH-values below 6.8
the nitrification rate decline quite notably, the rates could be as low as 10 to 20
percent of the rate at pH 7 when pH decreases to around 5.8 and 6. 90 percent of the
maximum rate is occurring at a pH of 7.5 and 8.5, and less than 50 percent of the rate
occurs below a pH of 6.4 and above 9.6 (Viessman, et al., 2009). However, values of
7.0 to 7.2 are often used to maintain a reasonable nitrification rate.
An optimal value of pH for denitrification ranges between 7.0 and 9.0 though local
conditions can cause variations. If the pH decreases below 7.0 it will affect the end
product since nitric oxides, such as , are produced when pH is declining (Henze,
et al., 2002). However, Viessman et al. (2009) claims the optimum pH to be between
6.5 and 7.5 (in accordance with m𝑁o2 s𝑂t heterotrophic bacteria) and that the rate is
decreased with up to 80 percent of the maximum if pH is reduced to 6.1 or raised to
7.9.
SLUDGE AGE
A research on nitrogen removal in wastewater using biological treatment showed that
the sludge age is a factor that affects the biological treatment. The nitrification process
was unstable in temperatures lower than 15°C and sludge age lower than 20 days.
Though, when the sludge age was higher than 20 days, the colder temperatures had
lower influence on the process which was considered to be stable. The sludge age
effect on the nitrification process was analyzed during winter (<15°C) and summer
(>15°C) conditions which resulted in that the sludge age had no influence on the
nitrification process during summer while a sludge age above 20 days eliminated the
negative effects of low temperatures on nitrification (Komorowska-Kaufman, et al.,
2006).
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
15 |
Chalmers University of Technology | 3 NITROGEN IN TUNNEL CONSTRUCTION
The drill and blast technique is used in Gerumstunneln which is a conventional tunnel
construction method used as the primary technique in Sweden (Vägverket, 2008).
The drill and blast technique is carried out in cycles where each cycle contains several
steps that are needed to blast a few meters of tunnel. The first step in the drill and
blast technique is most often pre-grouting. Groundwater leakages are often expected
during tunnel construction why pre-grouting will be conducted regularly. The next
step is drilling of boreholes in the required parts of the rock which are loaded with
explosives. The boreholes could be approximately 2-6 meters deep but vary
depending on the project (Vägverket, 2009). The third step is the actual blasting
followed by transportation of the blasted and crushed material. The last step is
mucking and scaling under which loose rock is removed at the blasted surface in order
to avoid any large blocks of rock to fall down. At times, extra reinforcement is
required such as bolting, shotcrete and post-grouting (Grinder, 2003). When one cycle
is performed, another cycle begins which is followed by more cycles until the
requested length of the tunnel is blasted. The cycles can vary in time due to variations
of the rock quality along the tunnel stretch.
3.1 EXPLOSIVES
Both drilling and blasting contribute to dust which is managed by flushing water on
the working face in order to settle the dust and reduce health risk for the workers.
Dust is thus transported via the water in the tunnel and could be a potential threat for
the recipient. The explosive itself during tunnel blasting is responsible for producing
several bi-products which also will be mixed with water, hence polluting the
wastewater further (Houfeng, et al., 2013). An explosive material can be defined as a
compound that reacts very rapidly and generates large quantities of gases in
conjunction with liberation of heat (Sudweeks, 1985).
The nitrogen residue from blasting is easily transferred into the water from the
working face. Sometimes, for practical reasons, not all explosives are fully detonated
and nitrogen in the chemical forms of ammonium and ammonia
contaminates water from remains of undetonated explosives (V+ikan & Meland, 2012)
and nitrates can leach from the explosives in wet blast𝑁 h𝐻o4 les or during char𝑁gi𝐻ng3
(Forsyth, et al., 1995). There are different types of explosives, but ammonium nitrate
is a common component. Nitrogen dioxide is produced with an ammonium
nitrate explosive which can be seen in Reaction 3.1 whilst Reaction 3.2 shows how
the gas , which is freely soluble, is
tran𝑁s𝑂fo2
rmed into nitric acid. Ammonium
nitrate is an emulsion explosive and the main pollutants are ammonia and nitrate
(Ho𝑁uf𝑂e2
ng, et al., 2013).
𝑁𝐻3
𝑁 𝑂3 (Re 3.1)
3 1
𝑁 𝐻4𝑁𝑂3 → 4𝐻2𝑂+2𝑁𝑂2
(Re 3.2)
1 2 1
𝑁Sp𝑂il2la+ge
3
o𝐻f2 e𝑂xp→lo 3si𝐻ve𝑁s𝑂 d3u+rin 3g𝑁 charging and undetonated explosives accounts for most
nitrogen that leaks to recipients near the construction. Blasted rock masses contain the
largest part of nitrogen compounds, as much as 60-70 percent of the nitrogen leaked
to water comes from these masses (Tilly, et al., 2006). Commercial and modern
explosives usually contain a fuel and oxidizer and examples of oxidizing agents are
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
16 |
Chalmers University of Technology | ammonium nitrate , calcium nitrate and sodium nitrate .
Explosives of common use can be divided into three groups; ANFO (ammonium
nitrate and fuel oil𝑁),𝐻 w4𝑁at𝑂e3 rgels/slurries and e𝐶m𝑎𝑁ul𝑂si3 ons. All groups of ex𝑁pl𝑎o𝑁si𝑂ve3 s
contain nitrogen compounds but with different water resistance, i.e. different
solubility in water, hence varying degrees of introducing nitrogen to water systems
(Forsyth, et al., 1995). Blasting work in tunnels generally requires more explosives
per m3 rock compared to open pits. Around 1.5-2 kg/m3 explosives per solid m3 rock
is used during tunnel construction in Sweden and around 50 000 ton of explosives are
used yearly in Sweden for mining, tunneling and other activities (Tilly, et al., 2006).
ANFO
The explosive ANFO is constituted by a relationship of around 95 percent ammonium
nitrate and 5 percent diesel oil. The nitrogen content is around 34 percent in ANFO
(Grinder, 2003) and the two nitrogen compounds ammonia and nitrate ions are very
water soluble forms of nitrogen. ANFO has no water resistance why nitrogen is
soluble when or if exposed to water. ANFO is also hygroscopic, i.e. absorbs any
available water and if ANFO absorbs too much water it becomes de-sensitized and
will not detonate (Forsyth, et al., 1995). This in turns leads to explosives being left
undetonated in boreholes, hence nitrogen could potentially leach to the construction
water.
SLURRY MIXTURES
Slurry mixtures, also known as water gels or dense blasting agents, consist of
sensitizer, an oxidizer, water and a thickener (Nichols Jr., 2005). Slurries contain the
same soluble nitrogen compounds as ANFO but with the difference that the water
resistance is good once the cross-linker has activated the gum. The gelled gum creates
an impermeable barrier between oxidizing agents and external water. The nitrogen
content is between 20 to 30 percent in the mixture according to Forsyth et al. (1995).
EMULSION
The third group of explosives is emulsion explosives (EMX) which as the two latter
groups also is constituted by the water soluble compounds ammonia and nitrate. The
water in oil emulsion is very resistant and there is a thin film of oil surrounding a salt
solution, which limits the contact with external water. An emulsion mixture contains
around 20 to 30 percent of nitrogen (Forsyth, et al., 1995). Emulsion explosives
containing no secondary explosives represent most commercial explosives produced
today. There are many advantages with emulsion explosives compared to other
explosives such as high safety in use, low cost, lower environmental impacts and the
possibility of production at the construction site thus reducing transports. The amount
of toxic nitrogen from emulsion explosives are 5-15 times less compared to
commercial explosives with ammonium nitrate with the secondary explosive TNT
(Yunoshev, et al., 2013). Emulsion explosives contain less pure nitrogen compared to
ANFO, around 33-75 percent of ANFO’s nitrogen content (Tilly, et al., 2006).
The explosive used in the Gerum tunnel is an emulsion explosive called site sensitised
emulsion (SSE) which become more popular in recent years due to developments of
using it underground. The explosive was first introduced by Dyno Nobel in 1995 for
underground blasting which is today known as the Titan SSE-system. This emulsion
matrix is classified as UN 5.1 oxidizing agent, meaning that is has other restrictions
regarding storage and transport compared to conventional explosives (Fauske, 2003).
SSE is an emulsion, hence the name, with an aqueous nitrate solution and an oil phase
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
17 |
Chalmers University of Technology | which is mixed with a chemical sensitizer (Bakke, et al., 2001). The matrix only
becomes explosive after a couple of minutes when it has been charged in the borehole
together with additive of gas. This system often include a charging truck, two
container tanks and pumps for the materials; emulsion matrix and sensitizing agent
(Fauske, 2003).
3.2 WATER SOURCES IN TUNNELING
There are several types of water sources during tunnel constructions that in different
ways are affected by the tunnel construction in terms of excessive nitrogen
concentrations. Some effects such as groundwater contamination can be crucial during
the project, but also long-term if the water management is not handled properly. This
section presents some different water sources in tunnel construction in order to clarify
how they differ amongst each other as well as their relevance in a tunnel construction.
BILGE WATER
During the construction process, inflowing groundwater and stormwater will be
collected in excavation shafts which can be contaminated by various pollutants from
the construction site (Trafikverket, 2011). This water accumulation is called bilge
water and need to be treated before reaching the recipient.
DRAINAGE WATER
Drainage water is groundwater that infiltrates in to the tunnel and needs to be
removed in separate drainage pipes (Trafikverket, 2011). The groundwater drops from
the sealed rock and are led to drainage pipes by a waterproof membrane. It can also be
led to the drainage pipes through the layer of macadam in the bottom of the tunnel
(Trafikverket, 2012).
LEACHATE WATER
Leachate is wastewater created by percolation of rainwater and moisture in landfills
through different types of waste. Leachate at tunnelling sites contains high
concentrations of ammonium nitrogen and organic and inorganic compounds are
transported from the waste to recipients (Hasar, et al., 2009). The rock material from
blasting has to be stored before it is crushed and used, thus storage areas at the project
site are exposed to explosive residues which contains nitrogen compounds.
Precipitation infiltrates the landfills and cause leachate water with high nitrogen
concentrations.
PROCESS WATER
During tunnel construction in rock, large volumes of process water are generated. The
process water is used during drilling of boreholes for blasting, probing and grouting,
as well as in cooling processes of the augers (drills). Large volumes of water are also
needed to moist blasted rock masses and to wash rock surfaces prior to using shotcrete
(Trafikverket & Ramböll, 2012). The process water is often contaminated due to
residues in the construction process, such as oil and explosives (Trafikverket, 2011).
Thus process water needs proper managing before discharged to recipients in order to
reduce the contamination risk to the environment. The process water from the Gerum
tunnel will be used in the lab analysis in Chapter 4 and consist of groundwater from
the area.
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
18 |
Chalmers University of Technology | 3.3 WATER TREATMENT IN THE GERUM TUNNEL
The first proposed water treatment process was introduced in August 2013 which
consisted of pH adjustment, a flocculation chamber, a lamella separator, DynaSand
Oxo (Figure 3.1) for nitrification, a carbon source and finally DynaSand filters
(Figure 3.2) with mechanical and contact filtration as well as biologically activated
filtration. But this treatment process failed to meet the demands from the Swedish
Traffic Administration, in terms of nitrogen concentration. Many efforts were done
trying to activate the biological treatment, but there were no successful outcomes,
which is assumed to be related to low temperatures (<10°C). The sludge was collected
Figure 3.1. Dyna Sand Figure 3.2. Lamella Filters (Nordic Water, 2014)
Filters (Nordic Water, 2014)
from a local wastewater treatment plant which could be, in combination with cold
water temperatures, one reason for why the nitrogen removal process decelerated.
Low water temperatures will be a reoccurring problem for tunnelling construction in
Sweden since the tunnel water will at most times be below 10°C (Hallberg, 2014).
Since the original water treatment design could not meet the demands of nitrogen
concentrations, especially the biological treatment, the treatment process was altered.
As from November 2013, the process water is now treated in fewer steps without the
biological treatment, hence the current treatment focuses on suspended particles, pH
and oil contaminants. This treatment is functioning as a pre-treatment for nitrogen
removal, which is considered to be efficient. The process water from the tunnel
construction is first led to a container with oil skimmer treatment and then via a 1 m3
container to a flocculation chamber. After the 1 m3container, a manometer is used
which is followed by addition of a carbon source if pH needs adjustment. The carbon
source is carbonic acid which needs proper mixing before continuing to the next step
which is additive of flocculants before the coagulation of particles in the flocculation
chamber. After flocculation, the water goes through a lamella separator to remove
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
19 |
Chalmers University of Technology | sludge. From the lamella separator, the water is transported to a new container before
ending up in an artificial dam and the sludge is collected by sludge trucks. The
process water is currently recycled back into the tunnel construction due to limited
water extraction from the water supply in the area which is around 1.8 m3 per hour.
Thus the artificial dam is used, which protects the recipient to some extent. However,
construction water outside the tunnel, e.g. rainwater mixed with blasting materials in
storage areas, is led to a natural dam before reaching the recipient. The natural dam
shows values precisely above allowed nitrogen concentrations. The main idea is to let
all water go back to the recipient but since the nitrogen levels have been too high
during tunnelling processes, it has to undergo recycling and is currently prohibited
from entering natural water sources.
Manometer Carbonic Acid
In-flow from Oil Separation 1 m3
tunnel Container
Suspended Lamell Flocculation
Particles/pH Filters Chamber
Flocculation pH
Agent
Artificial Sludge
Dam Removal
Figure 3.3. Schematic figure of the current water treatment plant for process water.
The current water treatment plant visualized in Figure 3.3 is nonetheless performing
efficiently and is not an issue since it keeps the previous mentioned parameters under
control. The problem is the absence of biological treatment to reduce nitrogen
compounds.
3.4 NITROGEN CONCENTRATIONS IN EFFLUENTS
There is limited information about nitrogen removal and nitrogen concentrations in
tunneling operations. However, the contaminated water i.e. the process water can be
compared to effluents in mining activities which is also polluted by excessive nitrogen
from explosives. There is more literature on mining, which will be used in order to
identify a typical range of nitrogen concentrations in effluents.
The Swedish mine company LKAB is used for reference values in terms of outgoing
process water. The values have been collected from 2011-2013 at two sites, Kiruna
and Vitåfors. The temperature is similar to that in tunnel water with a range of 0-19°C
and 1-21°C in Kiruna and Vitåfors with a median value of 3.7 and 7.5 respectively.
The range of ammonium nitrogen is 10 to 1 000 µg/l and the median value is 430 µg/l
at the site in Kiruna. The range in Vitåfors for ammonium nitrogen is 40 to 2 600 µg/l
and the median value is 964 µg/l. Thus, ammonium nitrogen is higher in Vitåfors. The
range for nitrate nitrogen is 2 100 to 28 900 µg/l and 2 000 to 49 800 µg/l in Kiruna
and Vitåfors respectively and the median for the two sites are 18 400 and 29 550 µg/l.
The range for total nitrogen varies between 25 700 to 43 900 µg/l with a median value
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
20 |
Chalmers University of Technology | of 35 450 µg/l at the site in Kiruna. The same parameter in Vitåfors has a range of
30 300 to 69 300 µg/l and a median value of 54 300 µg/l (Suup, 2014).
One research on biological removal of ammonia and nitrate in Ontario, Canada on
mine water had an inlet range of nitrate concentration between of 51 000-405 000 µg/l
(which converted to nitrate-nitrogen is 11 730-93 150 µg/l). This was considered to be
in the typical range of inlet nitrate concentrations. The general ammonia and nitrate
concentrations in mine effluents can vary quite considerably with 10 000-40 000 µg/l
for total ammonia and 25 000-300 000 µg/l for nitrate in the effluent water (Koren, et
al., 2000). These ranges converted to total ammonia-nitrogen and nitrate-nitrogen is
8 200 to 32 800 µg/l and 5 750-69 000 µg/l respectively. Poláková et al. (2013) used
mine water with ammonia-nitrogen concentrations of 10 000-11 000 µg/l and
achieved 86 percent removal efficiency. Ammonia is not monitored to the same extent
as ammonium, nitrate and nitrite, though it is interesting to know in which range
ammonia is usually present in mine- and tunnel water which can be used for
comparison. Other ammonia levels in mine water vary in the range of 500-26 800 µg/l
(410-21 976 µg/l ammonia-nitrogen) which is based on a project in Finland where
cold temperatures were analyzed in relation to the bacterial community. The nitrate
concentrations in this project varied between 7 200-52 500 µg/l (1 656-12 075 µg/l
nitrate-nitrogen) (Karkman, et al., 2011).
Another research on cold mine water was carried out in Finland in 2007 with water
from one gold mine and one chromite mine. This research showed that fixed-bed
biofilm reactors could be used to remove ammonium and nitrate in low temperatures.
The ammonium-nitrogen concentrations of the two dewatering systems were 24 200
and 17 700 µg/l respectively and 17 500 and 4 300 µg/l for nitrate-nitrogen
concentrations. However, ammonium rich water was presented as 50 000-100 000
µg/l why the ammonium concentrations in the two mines could be considered as
moderate or low (Zaitsev, et al., 2008).
According to Tilly et al. (2006), bilge water in tunnelling usually has a nitrogen
concentration of 100 000 to 1 000 000 µg/l compared to untreated wastewater with
20 000 to 40 000 µg/l. Landfill leachate could also be used as reference values for the
process water, however, this water has much higher nitrogen concentrations. The
composition of a landfill leachate, Tveta, in Södertälje, Sweden, had the range of
1 100 to 200 000 µg/l ammonium nitrogen between 1994 and 2002. The total nitrogen
concentrations during the same period had a range of 900 to 230 000 µg/l (Kietlinska,
2004).
The explosives SSE and ANFO were used in two different tunnels in the project
Botniabanan, performed by Skanska Sweden AB, in order to analyze the nitrogen
contaminants from different explosives. The Björnböle tunnel used SSE and had an
average nitrogen discharge in the process water of 170 000 µg/l but the levels varied
from 25 000 to 275 000 µg/l. The total amount of SSE was 573 kg with a specific
charge of 2.82 kg/m3 in the first analysis. The second analysis had a total charge of
620 kg SSE which gives a specific charge of 3.05 kg/m3. The Namntall tunnel used
ANFO as the primary explosive and the average nitrogen discharge in the process
water was 40 000 µg/l but it varied from around 10 000-50 000 µg/l. When the
handling of ANFO was improved, i.e. less ANFO outside the boreholes, the average
level decreased to 25 000 µg/l. The total charge was 290 kg of ANFO, DynoRex and
Dynocord which gives a specific charge of 2.1 kg/m3. Thus, the total use of
explosives is higher with SSE compared to ANFO. However, the nitrogen amount
discharge using SSE in relation to charged SSE in kilograms is half of that using
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
21 |
Chalmers University of Technology | 3.5 GENERAL DISCHARGE DEMANDS
Many municipalities in Sweden have similar restrictions to the most common
measured parameters in wastewater. Stockholm, Trollhättan and Malmö vicinity are
some examples of areas using guidelines based on Table 3.3 from Svenskt Vatten.
However, there could be local variances depending on external environmental factors.
Table 3.3 presents parameters that, according to Svenskt Vatten, could affect the
treatment processes and the water quality and are applied to industrial process water
(Svenskt Vatten, 2012).
Table 3.3. Discharge guidelines for some
parameters for industrial process water (Svenskt
Vatten, 2012)
Parameter Unit Level
pH - 6.5-10
Lead, Pb μg/l 50
Cadmium, Cd μg/l 0
Chromium, Cr μg/l 50
Nickel, Ni μg/l 50
Silver, Ag μg/l 50
Zinc, Zn mg/l 0.2
Ammonium mg/l 60
+
Oil index 𝑁𝐻4 mg/l 5-50
The Swedish Food Administration has set out some guidelines for many parameters in
drinking water. Table 3.4 presents the guidelines for when the water is unsuitable for
drinking water and the guidelines for when the water is potable.
Table 3.4. Guidelines from the Swedish Food Administration
(SLV, 2001)
Parameter Unit Limit unsuitable Limit potable
Cadmium μg/l 5.0
Copper mg/l 2.0 0.2
Chromium μg/l 50
Nickel μg/l 20
Nitrate mg/l 50 20
Nitrite mg/l 0.50 0.10
Ammonium mg/l 0.50
3.5.1 MUNICIPAL VARIATIONS
The municipality of Norrköping has provided guidelines for discharge of stormwater
to recipients of different sensitivities. These guidelines apply for stormwater discharge
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
23 |
Chalmers University of Technology | and are presented in Table 3.5. However, guidelines for process water from different
activities could have different and more restricted limits.
Table 3.5. Discharge guidelines from Norrköping municipality (Dagvattengruppen, 2009).
Discharge source
Parameter Unit From operational Subarea to To recipient with
activity recipient without protective values
protective values
Phosphorus, P mg/l 0.250 0.175 0.160
Nitrogen, N mg/l 3.5 2.5 2.0
Lead, Pb μg/l 15 10 8
Copper, Cu μg/l 40 30 18
Zinc, Zn μg/l 150 90 75
Cadmium, Cd μg/l 0.5 0.5 0.4
Chromium, Cr μg/l 25 15 10
Nickel, Ni μg/l 30 30 15
Suspended solids, SS mg/l 100 60 40
Oil index mg/l 1.0 0.7 0.4
The guidelines for discharge demands in Gothenburg are referred to the concentration
in the discharge point, i.e. the effluents. These guidelines have been developed using
several Swedish legislations and environmental goals and are seen in Table 3.6.
Table 3.6. Guidelines for effluent concentration (Göteborgs Stad Miljöförvaltningen,
2013)
Parameter Unit Concentration in effluents
Chromium, Cr μg/l 15
Cadmium, Cd μg/l 0.4
Lead, Pb μg/l 14
Copper, Cu μg/l 10
Zinc, Zn μg/l 30
Nickel, Ni μg/l 40
Oil index mg/l 1.0
pH - 6-9
Total phosphorus, TP μg/l 50
Total nitrogen, TN mg/l 1.25
TOC mg/l 12
Suspended solids, SS mg/l 25
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
24 |
Chalmers University of Technology | 3.5.2 DEMANDS IN GERUMSTUNNELN
Nitrogen concentrations are not always incorporated in the guidelines unless there is a
sensitive recipient. There are mainly two surface water sources near Gerumstunneln
which are Pulsebäcken and Gerumsälven. These two recipients are considered as
sensitive since the environmental assessment by the Swedish Nature Centre showed
presence of many different fish as well as beavers near Gerumsälven (Trafikverket &
Ramböll, 2012). An internal analysis of the two recipients showed increased nitrogen
concentrations but also increased metal concentration during the construction which is
related to the construction work. The Swedish Transport Administration set out
maximum concentration values for ammonium, nitrate and nitrite in Pulsebäcken and
Gerumsälven in Tanumshede prior to the tunnel project start-up. Table 3.7 presents
maximum concentration values for the three parameters which show that
Gerumsälven has higher nitrogen limits due to a higher flow.
Table 3.7. Acceptable parameter limits according to the STA
(Trafikverket & Ramböll, 2012)
Parameter Pulsebäcken Gerumsälven
Ammonium <78 µg/l <312 µg/l
Nitrate NH4−N <230 µg/l <1150 µg/l
Nitrite NO3−N <15 µg/l <30 µg/l
pH NO2−N 6-8 6-8
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
25 |
Chalmers University of Technology | 4 BIOLOGICAL NITROGEN REMOVAL TESTS
The aim with tests in the lab is to try to get the biological treatment to work in terms
of nitrification and denitrification of the process water. Nitrification and
denitrification tests using water with temperatures below 10°C will not be analyzed,
thus the water is at room temperature (19°). The reason for this is that the aim was to
first see if it is possible to conduct a complete nitrification test at all at room
temperature. The first experiment (Test 1) used only the process water and the
nitrifying and denitrifying sludge while the second test (Test 2) used, besides water
and sludge, sodium dihydrogen phosphate to compensate any potentially low
phosphate concentrations. The third nitrification test (Test 3) increased the phosphate
concentration as in Test 2 but the alkalinity was also increased. The reagents needed
for the test were process water, sludge and compressed air. Process water was
collected from two different days, 2014-03-31 and 2014-04-03. The water was stored
in a cool place and then brought to Chalmers. A denitrification test was also carried
out which used the process water from 2014-04-03 and a carbon source.
4.1 PREPARATIONS
Before starting the nitrification process, some preparations were needed. The sludge
concentration needed to be analyzed, the pH, the alkalinity and the nitrogen
concentrations of the process water and sludge liquor had to be known in order to
compare the results from the nitrification process, i.e. if there have been any activity
during the test. The pH was measured to 6.5 in the process water from 2014-03-31
and 6.4 in the water from 2014-04-03. Process water from the two different sampling
days was poured into three 50 ml test tubes each. These samples were frozen until
they could be analyzed in the ion chromatograph which will give the concentrations of
ammonium, nitrite and nitrate.
The alkalinity was decided for the two different samples of process water from 2014-
03-31 and 2014-04-03 by the use of titration. 50 ml water each from the two days was
used. Three drops of the indicator solution were added which turned the solution
turquoise (at a pH of 5.4). The aim of this method (Standard methods, EN ISO 9963-
1:1995) is to titrate the turquoise solution until it becomes transparent. The titration
solution used in this test was 0.02 M hydrochloric acid. The required volume of
hydrochloric acid in the water to obtain a transparent color was 5.4 ml for the process
water from 2014-03-31 and 5.3 ml for the water from 2014-04-03. When these
volumes have been determined, the alkalinity can be calculated using Equation 4.1.
The alkalinity for the two different water samples is calculated in Equation 4.2 and
4.3.
(Eq 4.1)
𝑐(𝐻𝐶𝑙)×𝑉5×1000
𝐴 = 𝑉4
𝐴 = 𝐴𝑙𝑘𝑎𝑙𝑖𝑛𝑖𝑡𝑦,𝑚𝑀/𝑙
𝑐(𝐻𝐶𝑙) = 𝐶𝑜𝑛𝑐𝑒𝑛𝑡𝑟𝑎𝑡𝑖𝑜𝑛 𝑜𝑓 𝑡ℎ𝑒 ℎ𝑦𝑑𝑟𝑜𝑐ℎ𝑙𝑜𝑟𝑖𝑐 𝑎𝑐𝑖𝑑,𝑀/𝑙
𝑉4 = 𝑉𝑜𝑙𝑢𝑚𝑒 𝑜𝑓 𝑡ℎ𝑒 𝑤𝑎𝑡𝑒𝑟 𝑠𝑎𝑚𝑝𝑙𝑒,𝑚 𝑙
𝑉5 = 𝑉𝑜𝑙𝑢𝑚𝑒 𝑜𝑓 𝑡ℎ𝑒 ℎ𝑦𝑑𝑟𝑜𝑐ℎ𝑙𝑜𝑟𝑖𝑐 𝑎𝑐𝑖𝑑 𝑐𝑜𝑛𝑠𝑢𝑚𝑒𝑑 𝑏𝑦 𝑡ℎ 𝑒 𝑠𝑎𝑚𝑝𝑙𝑒 ,(𝑚E q𝑙 4.2)
0.02×5.4×1000 𝑚𝑀
𝐴(2014/03/31) = 50 = 2.16 𝑙 ×61 = 131.76 𝑚𝑔/𝑙 (Eq 4.3)
0.02×5.3×1000 𝑚𝑀
𝐴(2014/04/03) = 50 = 2.12 𝑙 ×61 = 129.32 𝑚𝑔/𝑙
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
26 |
Chalmers University of Technology | The stoichiometry for nitrification and alkalinity is seen in Reaction 4.1
(Tchobanoglous, et al., 2004). For every mole of ammonium nitrogen to be
transformed to nitrate nitrogen, two moles of HCO are required.
3
(Re 4.1)
+ − −
However, for this test it is assumed that 2.5 mole of alkalinity is required per one
𝑁𝐻4 +2𝐻𝐶𝑂3 +2𝑂2 → 𝑁𝑂3 +2𝐶𝑂2+3𝐻2𝑂
mole ammonium nitrogen to make sure that the nitrification process is not limited by
alkalinity. This relation was used at Ryaverket in Gothenburg and was used to see if
the alkalinity was high enough for the ammonium nitrogen concentrations.
PROCESS WATER 2014-03-31
(Eq 4.4)
𝐴 = 2.16 𝑚𝑀
𝑚𝑔 104
𝑁𝐻4 −𝑁 = 104 𝑙 => 14 = 7. 43 𝑚𝑀 (Eq 4.5)
𝐴Th𝑟𝑒e𝑞 r =eq 7u .i 4r 3ed
×
a 2lk .5al =ini 1ty
8
.5f 8or
𝑚
a
𝑀
concentration of 104 mg/l is 18.58 mM
according to calculations in Equation 4.4 and 4.5, but the real alkalinity is only 2.16
mM. Thus, the alkalinity is too lo𝑁w𝐻 a4 n−d c𝑁ould inhibit the nitrification process.
PROCESS WATER 2014-04-03
(Eq 4.6)
𝐴 = 2.12 𝑚𝑀
𝑚𝑔 111
𝑁𝐻4 −𝑁 = 111 𝑙 => 14 = 7. 93 𝑚𝑀 (Eq 4.7)
𝐴Th𝑟𝑒e𝑞 a =lka 7l .i 9n 3ity
×
w 2a .5s =cal 1c 9u .l 8a 3te 𝑚d 𝑀for this process water in Equation 4.6 which was 7.93
mM. The calculated required alkalinity in Equation 4.7 was 19.83 mM, thus the
alkalinity in the water is too low in this sample which could limit the nitrification
activity.
SUPERNATANT 2014-05-02
The supernatant is the top layer liquid produced when the bigger particles of the
sludge is allowed to settle. The alkalinity for the supernatant from May 2 was
determined using the same method as for the process waters with a 50 ml sample and
a 0.02 M acid. The required volume of the hydrochloric acid to make the sample go
from turquoise to transparent was 5.8 ml. The alkalinity of the sludge is seen in
Equation 4.8:
(Eq 4.8)
0.02×5.8×1000 𝑚𝑀
𝐴4(.220 14/N05I/T0R2)I=FICAT50ION T=E2.S3T2 1𝑙 ×61 = 141.52 𝑚𝑔/𝑙
The first nitrification test was conducted on April 7 2014 where 1 500 ml samples of
process water from the two days were mixed with 500 ml sludge using a magnetic
stirrer. The solution was continuously aerated throughout the experiment. A total
number of 26 samples, each 20 ml in size, were collected from the two solutions
according to the schedule in Table 4.1. Each 20 ml sample was first centrifuged 4 000
RCF for 5 min. Thereafter the samples were filtered through 0.45 µm and stored in
the refrigerator. These samples were later frozen for analysis due to maintenance
problems of the ion chromatograph. The temperature was around 19°C in the two
containers during the nitrification test.
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
27 |
Chalmers University of Technology | and with 31 samples, including two tubes of MQ water, the analysis was run for
approximately 10 hours. However, there were certain preparations required before
using the ion chromatograph. The 26 samples from the nitrification test were diluted 5
times with 2 ml of the sample and 8 ml MQ water. Three more samples were run in
the chromatograph, the filtered samples of the process water and the sludge which
gives a total of 29 samples. The first and last sample in the ion chromatograph is
always filled with only MQ water which is used as a calibration to see that the
instrument is working properly.
4.2.1 SLUDGE CONCENTRATION TEST 1
The sludge was collected from Hammargårdens treatment plant in Kungsbacka on
April 7 (for Test 1 and 2) and May 2 (for Test 3 and the denitrification). This sludge
consists of both nitrifying and denitrifying bacteria which are required for a
nitrification and denitrification processes. The sludge concentration was
approximately 4.6-5 g/m3 or 4 600-5 000 mg/l on April 7 and 2.3 mg/l on May 2
according to the instrument at site , however, the concentration was measured during
the lab at Chalmers as well to obtain a more accurate concentration.
First, the sludge was allowed to settle for a few minutes in order to remove most of
the supernatant. 5 ml samples were collected of the sludge before weighing 1.6 µm
filters which was required for the calculations of sludge concentrations. The 5 ml
samples were then filtered through the 1.6 µm filters after which the filters were
heated in the oven in 105°C for 2.5 hours for filter 1 and 2 (the minimum time for
heating is 2 hours). The filters were cooled before being weighed a second time. The
sludge sample from Test 2 was diluted with the process water, thus lower
concentration. The filters were then heated in the oven again, but this time in 550°C
for 30 min in order to measure the organic material content. Suspended solids could
be calculated if the weight of the filter before and after heating in 105°C is known.
The following equation is used:
(Eq 4.9)
𝑋𝑑𝑟𝑦−𝑋𝑤𝑒𝑡
𝑣
𝑋𝑑𝑟𝑦 = 𝑡ℎ𝑒 𝑤𝑒𝑖𝑓𝑔ℎ𝑡 𝑜𝑓 𝑡ℎ𝑒 𝑓𝑖𝑙𝑡𝑒𝑟 𝑎𝑓𝑡𝑒𝑟 2 ℎ𝑜𝑢𝑟𝑠 ℎ 𝑒𝑎𝑡𝑖𝑛𝑔 𝑖𝑛 105°
, l
𝑋𝑤𝑒𝑡 = 𝑡ℎ𝑒 𝑤𝑒𝑖𝑓𝑔ℎ𝑡 𝑜𝑓 𝑡ℎ𝑒 𝑓𝑖𝑙𝑡𝑒𝑟 𝑏𝑒𝑓𝑜𝑟𝑒 ℎ𝑒𝑎𝑡𝑖𝑛𝑔
The weight of the filters and the suspended solids are presented accordingly:
𝑣 = 𝑣𝑜𝑙𝑢𝑚𝑒 𝑜𝑓 𝑡ℎ𝑒 𝑠𝑎𝑚𝑝𝑙𝑒
Filter 1
• Before usage 0.0897 g
• After heating in 105°C 0.1241 g
• After heating in 550°C 0.0946 g
Suspended solids
0.1241−0.0897 𝑔 6.88
Filter 2 0.005 = 6.88 𝑙 => 4 = 1.72 𝑔/𝑙
• Before usage 0.0888 g
• After heating in 105°C 0.1249 g
• After heating in 550°C 0.0941 g
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
29 |
Chalmers University of Technology | Suspended solids
0.1249−0.0888 𝑔 7.22
The nitrification rate 0s .0c 0a 5n no=w 7b.2e 2c 𝑙al=cu>late 4d =for1 .t8h1e 𝑔tw/𝑙o filters since the sludge
concentrations is known. However, since this test failed to achieve a nitrification
process, the rates will not be relevant.
A 50 ml sample of the supernatant was also filtered through 1.6 µm on April 7 which
was frozen for further analysis of the sludge content. The pH of the sludge, measured
in the supernatant on April 7, was 6.84. The sludge age is not known but was
approximated to be between 15-20 days according to the staff at Hammargården.
4.2.2 RESULTS TEST 1
The results from the ion chromatograph fluctuated and did not present values for all
samples. The ion chromatograph turned out to give invalid results, this was
determined after the instrument was run with some standards solutions. These samples
were however kept and frozen until a new analysis was carried out when the
instrument was repaired. A second analysis with the samples from April 7 was run on
April 15 under correct conditions of the instrument. The concentration of phosphate
was insignificant and below the detection limit, which could be the reason for why the
biological treatment failed at site in Tanumshede.
The theoretical sample 0 (start concentration after mixing process water with activated
sludge suspension) is calculated to compare with the results from the ion
chromatograph using Equation 4.10.
(Eq 4.10)
1.5 ×𝑃 +0.5×𝑆
𝑆 𝑎𝑚𝑝𝑙𝑒 0𝑋 = 2
𝑆𝑎𝑚𝑝𝑙𝑒 0𝑋 = 𝑡ℎ𝑒 𝑓𝑖𝑟𝑠𝑡 𝑠𝑎𝑚𝑝𝑙𝑒 𝑖𝑛 𝑡ℎ𝑒 𝑛𝑖𝑡𝑟𝑖𝑓𝑖𝑐𝑎𝑡𝑖𝑜𝑛 𝑡𝑒𝑠𝑡
𝑃 = 𝑐𝑜𝑛𝑐𝑒𝑛𝑡𝑟𝑎𝑡𝑖𝑜𝑛 𝑖𝑛 𝑡ℎ𝑒 𝑝𝑟𝑜𝑐𝑒𝑠𝑠 𝑤𝑎𝑡𝑒𝑟,𝑚𝑔/𝑙
The theoretical samples for ammonium, nitrite and nitrate are calculated for the
𝑆 = 𝑐𝑜𝑛𝑐𝑒𝑛𝑡𝑟𝑎𝑡𝑖𝑜𝑛 𝑖𝑛 𝑡ℎ𝑒 𝑠𝑢𝑝𝑒𝑟𝑛𝑎𝑡𝑎𝑛𝑡,𝑚𝑔/𝑙
process waters from 2014-03-31.
1.5 ×104 +0.5×5
𝑆𝑎𝑚𝑝𝑙𝑒 01 (𝑁𝐻4 −𝑁) = = 79.3 𝑚𝑔/𝑙
2
1.5 ×5.6 +0.5×1.6
𝑆𝑎𝑚𝑝𝑙𝑒 01 (𝑁𝑂2 −𝑁) = = 5.4 𝑚𝑔/𝑙
2
1.5 ×346 +0.5×2.9
W𝑆𝑎h𝑚en𝑝 𝑙c𝑒o 0m1p (a𝑁ri𝑂ng3 −the𝑁 )th=eoretical values with the a=na2ly6z1e.d7 𝑚va𝑔lu/e𝑙s , 97.2, 5 and 305.6
2
mg/l respectively, it is clear that there have been some dilution faults. The theoretical
first values of the samples using process water from 2014-04-03 are also calculated.
1.5 ×111 +0.5×5
𝑆𝑎𝑚𝑝𝑙𝑒 01 (𝑁𝐻4 −𝑁) = = 84.5 𝑚𝑔/𝑙
2
1.5 ×6 +0.5×1.6
𝑆𝑎𝑚𝑝𝑙𝑒 01 (𝑁𝑂2 −𝑁) = = 4.9 𝑚𝑔/𝑙
2
1.5 ×386 +0.5×2.9
𝑆𝑎𝑚𝑝𝑙𝑒 01 (𝑁𝑂3 −𝑁) = = 290.2 𝑚𝑔/𝑙
2
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
30 |
Chalmers University of Technology | The values from the ion chromatograph show 71.7, 4.6 and 191.2 mg/l respectively
which also indicate that there are some errors with the dilution. The results from the
ion chromatograph were therefore normalized against sodium (anion) and chloride
(cation) to get more accurate results. These compounds are expected to be constant
during the experiment and by normalizing the data with these concentrations the
errors in the measurements should be eliminated. Normalization using sodium had the
best correlation and is presented in Figure 4.2.
a) d)
b) e)
c) f)
Figure 4.2. Results from Nitrification Test 1. Normalized values in figures a-c show how well sodium
is correlating with the compounds. Figures d-f show the trend for ammonium, nitrite and nitrate.
T he normalization was carried out by analyzing the ammonium, nitrite and nitrate
concentrations in relation to sodium and chloride in order to consider the potential
dilution errors. Thus the zero sample of the sodium concentration of the process water
was used for samples when normalizing with sodium. The first (sample zero) was
divided with the sodium concentration of the next sample X and then multiplied with
n
the ammonium concentration of that same sample, seen in Equation 4.11. The first
sodium sample is then divided with the third sodium sample X and multiplied with
n+1
the third ammonium sample. The same procedure is performed with nitrite and nitrate
and with chloride as the normalizing agent. Sodium and ammonium is thereafter
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
31 |
Chalmers University of Technology | Nitrification Test 1
TOC
10
8
6
31-mar
4
03-apr
2
0
Time (hours)
0 5 10 15 20 25
Figure 4.4. Results TOC Nitrification Test 1
The concentrations of TN are presented in Figure 4.5. The concentration of TN should
be on a similar level throughout a nitrification test since ammonium should be
transformed into nitrate. However, in Test 1 the concentration of TN varied with a
slight trend of increasing which could be due to disintegration of sludge flocs during
the experiment.
Nitrification Test 1
TN
65
55
45
31-mar
35 03-apr
25
Time (hours)
0 5 10 15 20 25
Figure 5.5. Results TN Nitrification test 1
4.3 NITRIFICATION TEST 2
This nitrification test was carried out on April 10 using the same method as Test 1.
However in this test, 0.078 g sodium dihydrogen phosphate was added due to the low
phosphate concentrations in Test 1 (which were below detection limit). This gives a
phosphorus concentration of 10 mg/l and could help the nitrification since there
required phosphate concentration could be too low. Samples were taken out according
to the schedule in Table 4.2 and the pH was measured. All samples were frozen after
Test 2 for further analysis. The temperature was around 16-17°C throughout the
experiment.
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
33
)l/gm(
NT
)l/gm(
COT |
Chalmers University of Technology | Table 4.2. Sampling scheme for nitrification Test 2
Time Process water Process water
2014-03-31 2014-04-03
Min pH Sample pH Sample
0 6.9 0 6.84 0
15 - 1 - 1
30 - 2 - 2
45 - 3 - 3
60 8.31 4 8.25 4
90 - 5 - 5
140 - 6 - 6
170 - 7 - 7
220 8.31 8 8.25 8
1030 8.01 9 6.56 9
4.3.1 SLUDGE CONCENTRATION TEST 2
The sludge concentration in Test 2 was determined by the same method as for Test 1
with the only difference that the 5 ml sample was taken from the mixed solution in
this test (process water and sludge). Filters 3 and 4 were heated in the 105°C oven
around 14 hours (they were kept in the oven overnight) and for 30 min in the 550°C
oven.
Filter 3
• Before usage 0.0908 g
• After heating in 105°C 0.1020 g
• After heating in 550°C 0.0936 g
Suspended solids
0.1020−0.0908
Filter 4 0.005 = 2.24 𝑔/𝑙
• Before usage 0.0878 g
• After heating in 105°C 0.0962 g
• After heating in 550°C 0.0885 g
Suspended solids
0.0962−0.0878
4.3.2 RES0.U00L5TS T=E1S.6T8 2𝑔 /𝑙
The ion chromatograph was used on April 22 for the samples from the Nitrification
Test 2. The theoretical sample 0 for ammonium, nitrite and nitrate are calculated for
the process waters from 2014-03-31 using the same equation as in Section 4.2.2. The
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
34 |
Chalmers University of Technology | TOC is still low in concentration and varies throughout the test. TN seems to first
decrease before it increases again, thus there is no complete nitrification in this test as
already declared.
Nitrification Test 2
TN
120
100
80
60
31-mar
40
03-apr
20
0
Time (hours)
0 5 10 15 20
Figure 4.7 Results TN Nitrification Test 2
4.4 NITRIFICATION TEST 3
Since neither Test 1 nor Test 2 resulted in any evident nitrification activity, another
test was carried out, Test 3. It was suspected from the alkalinity tests that the
alkalinity could be to low which was considered in this test. The proportion of the
process water and sludge was altered in this test to see if that could have been a factor.
The solution for the nitrification consisted of 1 000 ml process water and 1 000 ml
sludge. The phosphate was, as in Test 2, increased by adding 0.078 g sodium
dihydrogen phosphate in each container. The alkalinity was increased in this test by
adding 1.34 g sodium hydrogen carbonate which was determined by looking at the
ammonium concentration of the process water and the stoichiometry of alkalinity
(estimated that one mole of ammonium nitrified consumes two moles of alkalinity).
The following calculations were made in order to determine the required volume of
sodium hydrogen carbonate to increase the alkalinity to meet the
ammonium nitrogen concentrations:
𝑁𝑎𝐻𝐶𝑂3
• Process water March 31
𝑚𝑔
𝑁𝐻4−𝑁 104 52 𝑙
2• =Pro2ces=s w1a4ter= A3pr.7il 𝑚3 𝑚𝑜𝑙𝑒 3.7×2 = 7.4 𝑚𝑚𝑜𝑙
𝑚𝑔
𝑁𝐻4 −𝑁 111 55.5
𝑙
The test w=as run =for 22 hour=s a4n 𝑚d 𝑚15𝑜 𝑙m𝑒i n4 w×it2h =a 8to 𝑚tal𝑚 o𝑜f𝑙 12 samples per container.
2 2 14
Samples were collected at continuous intervals presented in Table 4.3 with the
majority of the samples collected during the first 5 hours.
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
37
)l/gm(
NT |
Chalmers University of Technology | Table 4.3. Sampling scheme nitrification Test 3
Process water Process water
Time 2014-03-31 2014-04-03
Min pH Sample pH Sample
0 7.83 0 7.82 0
15 - 1 - 1
30 - 2 - 2
45 - 3 - 3
60 8.29 4 8.16 4
90 - 5 - 5
120 - 6 - 6
150 - 7 - 7
180 8.26 8 7.98 8
240 - 9 - 9
300 8.14 10 7.85 10
1335 7.79 11 7.97 11
4.4.1 SLUDGE CONCENTRATION TEST 3
The sludge concentration was determined with the same method that was used in Test
1. The 5 ml sample that was filtered was taken from the sludge solution after it had
been thickened i.e. much of the supernatant had been removed. Filters 5 and 6 were
heated for 2.5 hours in 105°C and 30 min in 550°C.
Filter 5
• Before usage 0.0949 g
• After heating in 105°C 0.1505 g
• After heating in 550°C 0.0885 g
Suspended solids:
0.1505−0.0949 𝑔 11.12
Filter 6 0.005 = 11.12 𝑙 => 4 = 2.78 𝑔/𝑙
• Before usage 0.0926 g
• After heating in 105°C 0.1424 g
• After heating in 550°C 0.1003 g
Suspended solids:
0.1424−0.0926 𝑔 9.96
The nitrification ra te c 0a .n
00
b 5e cal=cul9a.t9e6d 𝑙us=in>g th 4e s=lu2d.g4e9 c 𝑔o/n𝑙centration in Section 4.2.1
and Equation 4.12. However, since the nitrification only worked from ammonium to
nitrite, which is the only nitrification rate that is possible to obtain. These calculations
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
38 |
Chalmers University of Technology | undetected during the analysis, the additive of sodium hydrogen carbonate could be
incorrect in order to achieve the wanted alkalinity.
Test 3 was also analyzed for TOC and TN. The result for TOC is presented in Figure
4.10 for the two different types of process water. The figure shows an increasing
curve of TOC concentration, though; there are some negative values for the process
water from April 3. This could be a result of the concentration being very close to
detection limit, thus limits that the machine is unable to discover. Therefore the
machine could have resulted with some negative values in these measurements.
Nitrification Test 3
TOC
4
3
2
1 31-mar
0 03-apr
0 5 10 15 20 25
-1
Time (hours)
-2
Figure 4.10. Results TOC nitrification test 3
Figure 4.11 shows the trend for TN for both process waters. The results for the two
days are very similar and show that TN is staying at the same level, which indicates
that some nitrification might be occurring since ammonium turned into nitrite.
Nitrification Test 3
TN
50
40
30
20 31-mar
10 03-apr
0
Time (hours)
0 5 10 15 20 25
Figure 4.11. Results TN Nitrification Test 3
4.5 DENITRIFICATION TEST
A denitrification test was also carried out in the lab at Chalmers University on May 5
using sludge from May 2 collected at Hammargården wastewater treatment plant. The
process water from April 3 was the only type used in this test due to the similarities in
water compositions between March 31 and April 3. April 3 had slightly higher
concentrations of nitrate why this sample was chosen. The sludge was first thickened
by removing some of the supernatant. It was also aerated for 30 minutes before used
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
41
)l/gm(
COT
)l/gm(
NT |
Chalmers University of Technology | in the experiment since it had been stored cool for a few days. 1 100 ml of water and
900 ml sludge were first mixed by a magnetic stirrer. Nitrogen gas was then added for
30 min before the carbon source was added. The carbon source was acetate
and the required amount was calculated to 2.86 g/2 l using the stoichiometry fo−r
acetate and the concentration of nitrate which was determined to 386 mg𝐶/l𝐻 i3 n𝐶 𝑂th2 e
nitrification test. The experiment was run for 3 hours (based on other standard
denitrification tests) during regularly sampling, which is seen in Table 4.5. Sample 0
A
is when acetate was added but due to measurement errors it was only 1 900 ml in total
why another 100 ml of process water was added when sample 0 was taken. The
B
samples were centrifuged for 5 min as in the nitrification tests and then filtered
directly through a 0.45 µm filter. The nitrogen samples were diluted 5 times using 2
ml filtered samples and 8 ml MQ water. These samples were then run in the ion
chromatograph.
Table 4.5. Sampling schem e for
denitrification test
Process water
Time 2014-04-03
Min pH Sample
0 - 0
A
8 - 0
B
15 7.97 1
30 - 2
45 - 3
60 - 4
90 8.29 5
120 - 6
150 - 7
180 8.47 8
4.5.1 SLUDGE CONCENTRATION DENITRIFICATION
The sludge concentration in the denitrification test was also determined by using the
same method as described in Test 1, Section 4.2.1. Filters 7 and 8 were heated for 2.5
hours in 105°C and 30 min in 550°C.
Filter 7
• Before usage 0.0954 g
• After heating in 105°C 0.1431 g
• After heating in 550°C 0.1017 g
Suspended solids:
0.1431−0.0954 𝑔 9.54
Filter 8 0.005 = 9.54 𝑙 => 4 = 2.39 𝑔/𝑙
• Before usage 0.0934 g
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
42 |
Chalmers University of Technology | • After heating in 105°C 0.1406 g
• After heating in 550°C 0.0999 g
Suspended solids:
0.1406−0.0934 𝑔 9.44
4.5.2 RES0U.00L5TS D=E9N.4IT4R𝑙 =IF>ICA4 T=IO2N.3 6T 𝑔E/S𝑙T
Figure 4.12 presents the results from the denitrification test. The denitrification
process should give decrease in nitrate before transformed to nitrite which in turn
should be transformed to nitrogen gas. The results show increasing nitrite and
decreasing nitrate which indicates that the denitrification process was working to
some extent. However, the test was only conducted during 3 hours at which the nitrate
concentration was measured to around 110 mg/l, which is still a high concentration
since the goal was to decrease nitrate close to zero mg/l. Thus the nitrate
concentration was decreased around 20 mg/l while nitrite was increased with 40 mg/l.
This shows that the denitrification was not fully complete since not all nitrate was
transformed into nitrite.
The pipette was working when the samples were diluted which would make the
results valid. But the results were normalized as in Test 3 with sodium and chloride.
a)
b)
c)
Figure 4.12. Results denitrification test
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
43 |
Chalmers University of Technology | 5 DISCUSSION
The main purpose of this master thesis was to conduct nitrification and denitrification
tests in the lab with process water from Gerumstunneln. The concentrations of
ammonium, nitrite and nitrate were measured in the first test and the other tests were
based on these concentrations. Though, when the process water was analyzed in the
third nitrification test presented in Section 4.4.2, the nitrogen concentrations had
increased. This could have affected the results in the test, in particular the calculations
used to increase alkalinity in Test 3 since it was based on the nitrate concentrations
from Test 1.
5.1 NITRIFICATION TESTS
The results showed a small decrease in ammonium (which should occur), a very small
increase in nitrite and finally fluctuating values for nitrate. This shows that there was
no complete nitrification, nor a very good result for the start of the nitrification
process. It was suspected that the phosphate concentration was too low in the process
water why it was decided to increase the concentration in the next nitrification test. It
was also believed that the dilution had not been performed with great accuracy due to
the pipette used. Thus, the results were normalized using sodium (cation) and chloride
(anion) and the theoretical first value was calculated. The theoretical values were not
consistent with the measured values why the conclusion was that the dilution had been
compromised during the use of the pipette, where the volume in the test tubes varied
from around 9.5-10.7 ml, in which there should have been exactly 10 ml. The
correlation between sodium and ammonium, nitrite and nitrate were very good. The
normalization with chloride also showed good correlation but sodium proved to be the
best one why the diagrams and results are based on sodium. Since the results showed
no clear sign of nitrification, a new test needed to be carried out. The concentration of
TN varied during Test 1, which should have been at the same level throughout the
entire test if the nitrification had been completed since ammonium is transformed into
nitrate, thus the amount of TN should still be the same. But with the result presented
in Section 4.2.2, it shows that the process was not working properly. The reason for
both increases and decreases of TN during the test could be the lack of phosphate, low
alkalinity and possibly lack of aeration during certain periods due to irregularly
pressure in the flow. It could also have been simultaneous denitrification due to the
irregular flow if the aeration was insufficient.
The second nitrification test used the same process water as in Test 1. Since the
results in Test 1 did not show nitrification activity as hoped, it was decided to increase
the phosphate concentration to see if it would help the process. These samples were
diluted at the same time as Test 1 (the samples had been frozen) which gave Test 2 the
same dilution errors as Test 1 since the same pipette was used. The results were
therefore normalized with sodium and chloride to see if any of the two had better
correlation. Sodium proved to have better correlation again and these values were
used when analyzing the curves and ammonium, nitrite and nitrate were plotted
against the time. Ammonium should decrease during nitrification, and since
phosphate had increased the hope was to achieve better results. However, the trend for
ammonium or nitrite was not very distinctive. This shows that nitrification was not
working in this test either, despite increase of phosphate. The pH was measured in the
beginning of the test to around 6.9 for both process water solutions, but the process
water from March 31 increased during the test to 8.3 after 220 min but stopped at 8
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
45 |
Chalmers University of Technology | after the entire test. The process water from April 3 also reached around 8.3 during the
test but when the pH was measured after 17 hours, it had decreased to 6.6. The pH
should be between 6.5 and 8.0 for nitrification to occur, which the pH was during the
test, thus the pH is most likely not the parameter causing nitrification failure. The
concentration of TN during Test 2 fluctuated, as in Test 1, which further proves that
this test had an incomplete nitrification.
The third nitrification test used sludge collected the same day from the wastewater
treatment plant. From the previous two tests, and since something in the project was
decreasing the alkalinity in the recipients, it was suspected that the alkalinity was too
low. The alkalinity was determined to 2.16 mM and 2.12 mM for the process waters
which were below the required alkalinity for the high concentration of ammonium in
the waters. The alkalinity was therefore increased by adding sodium hydrogen
carbonate. The phosphate concentrations were increased to the same levels as in Test
2. The pH was monitored and it was around 7.8 in the beginning of the test for both
process waters. The process water from March 31 increased to 8.3 but decreased to
7.8 at the end of the test. The process water from April 3 also increased in the middle
of the test to 8.2 and then decreased to 8 at the end. The pH has been in the range of
what is preferable for nitrification (6.5-8.0) which should not have impacted the
nitrifying activity. If pH increases in the solution, thus becoming more basic,
ammonium can transform into free ammonia. However, since pH has been kept at a
stable level, this should not be the reason for nitrification failure. The pipette had
undergone maintenance and gave good results without normalizing with sodium and
chloride. Nonetheless, the final step in the nitrification process i.e. nitrite to nitrate,
did not work in this test either. The nitrate concentrations were kept at a similar level,
thus an incomplete nitrification. The ammonium levels were close to zero mg/l at the
end of the nitrification test but if the test would have been kept running longer, it is a
possibility that all ammonium would have transformed into nitrite which maybe
would have started to transform into nitrate. The concentrations of TN during Test 3
remained at the same levels which indicate that some nitrification is occurring since
the amount of nitrogen compounds is similar during the entire test with the exception
of a small increase at the end of the test. Thus, nitrification was occurring at the
beginning of the test but was inhibited between the nitrite-nitrate transformations.
5.2 DENITRIFICATION TEST
There were only one denitrification test performed in this study. Denitrification
requires a carbon source and acetate was used for which the amount was determined
based on the nitrate concentrations from Test 1. The phosphate and the alkalinity
concentrations were not increased in this test in order to observe the performance of
only process water, sludge and acetate. The aim with denitrification is to change
nitrate into nitrite which turns into nitrogen gas. The dilution for this test was also
done accurately with the pipette and should not need normalization. The results show
that nitrate is decreasing while nitrite is increasing. Ammonium is also decreasing but
with some fluctuating values. The results are quite good for the denitrification, even
though not all ammonium is removed. The test was only run for 3 hours, which
decreased nitrate from 130 mg/l to 110 mg/l (Δ 20 mg/l) while nitrite increased from
10 mg/l to 50 mg/l (Δ 40 mg/l). If the test would have been run for longer time, e.g.
around 20 hours as in the nitrification tests, the nitrate could have been further
decreased and nitrite further increased. Denitrification aims to decrease nitrogen
levels and the concentration of TN shows a decrease during the test. This proves that
some of the nitrate and nitrite are transformed into nitrogen gas which has left the
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
46 |
Chalmers University of Technology | solution, thus denitrification is partially in progress. However, when assessing the
results it is seen in Figure 4.12 that all nitrate is not transformed into nitrite and all
nitrite is not transformed into nitrogen gas. This test had other sludge-process water
proportions compared to the nitrification tests, i.e. more sludge was used in relation to
the volume of water. This could have contributed to better results and it would have
been interesting to run a test with 50/50 proportions for all three nitrification tests as
well to see if the results would be different. If phosphate and alkalinity concentrations
would have been increased in this test it might have resulted in an even better
outcome since it already showed some denitrifying activity.
5.3 SLUDGE
The sludge used was collected from a treatment plant in Kungsbacka which had both
nitrifying and denitrifying bacteria. Due to logistics, the sludge was only used fresh
i.e. the day it was collected from the plant, for Test 1 and Test 3. For Test 2 and the
denitrification test, the sludge had been kept cooled in containers for maximum 3
days. This could affect the processes even though the sludge was aerated before
usage. It would have been preferable and probably better for the tests to have new
daily sludge for all tests, but due to logistic difficulties of collecting the sludge, the
same sludge was used in more than one test. However, another aspect of failed
nitrogen removal could be related to the sludge in terms of acclimation to the
environments for the bacteria. The bacteria in the sludge are not used to process water
from a tunnel that was used in this study, i.e. those bacteria are used to regular
wastewater. Bacteria sometimes need time to acclimate to new environments and time
for acclimatization was not provided for the bacteria prior the tests. This could be a
factor that should be assessed further, whether or not the bacteria in the sludge need to
be familiar with the process water before running a nitrification and denitrification
test. For further analysis, it would be interesting to run nitrification tests using the
same process water but different sludge from different wastewater treatment plants to
see if the processes differ or if the problem with nitrification only relates to the
composition of the water. The sludge age could also be a factor inhibiting
nitrification. The exact sludge age was never known, but an approximated guess at the
wastewater treatment plant was 15-20 days. It would have been interesting to use
sludge with higher age since sludge older than 20 days has proved to help the
nitrification process in cold temperatures.
5.4 SUMMARY
No tests in this analysis showed any distinct results on nitrification. They were all
incomplete and the nitrification process only went from ammonium to nitrite, but it
failed to change into nitrate which was the common factor for all three nitrification
tests. Thus, the second step in the nitrification test is causing the failure and the reason
for this is unclear. The denitrification process was quite successful since nitrate
decreased, nitrite increased and TN decreased. However, since the total time was not
sufficient enough to remove larger quantities of nitrogen, the test should have been
performed much longer. The reason for why nitrification failed is still undetermined.
Some thoughts and assumptions have been developed along the analysis. First, the
concentrations of nitrogen in the process water are much higher than in other
activities, such as in mining industries or in regular wastewater, which increases the
pressure on the biological treatment. Some metals have inhibiting effects on
nitrification such as zinc, copper and nickel, and some of these metals had increased
in concentration in both Pulsebäcken and Gerumsälven during the tunnel project. The
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
47 |
Chalmers University of Technology | metal increase comes most probably from the process water in the tunnel. The impacts
from metals have not been taken into account during the lab. This could therefore be a
reason, in combination with other factors such as phosphate and alkalinity, of the
nitrification failure. It is a possibility that the phosphate concentrations should have
been increased even further in Test 2 and Test 3 in order to achieve a complete
nitrification since the same concentration was used for the two. Same connection
applies for alkalinity which could have been increased even more in Test 3. Other
inhibiting parameters are pH and nitrite itself. However, pH has been quite stable
during the tests and has almost at all times kept within the range of 6.5-8.0 and should
not be the primary affecting parameter. The concentration of nitrite could also inhibit
the process if it is too high but nitrite was below the critical levels. Another problem
could also be a shortage of micronutrients that are necessary for microbial growth.
The problem for projects such as Gerumstunneln is the cold water in the biological
treatment which inhibits the nitrification process in terms of decreasing the
nitrification rate. However, all tests in the lab were performed in room temperature
and the nitrification process was still not completed. This indicates that the low
temperature is not the major issue, even though it could be a problem as well. It was
assumed prior to this study that nitrification would be possible in at least room
temperature and that the problem would be to manage the cold water. The analysis in
the lab does however present other complex results which brings more questions than
just those about low temperatures. There is no clear answer to why the nitrification
tests were uncompleted, but it is not only cold water that contributes to the failure, the
proportions of sludge and process water and the concentrations of metals could also
have been a strong factor to the results.
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
48 |
Chalmers University of Technology | 6 CONCLUSION
The conclusions of this master thesis are stated in this section.
• The first nitrification test resulted in an incomplete nitrification which was
assumed to be correlated with low phosphate concentrations in the process
water.
• The second nitrification test also showed incomplete nitrification despite a
phosphate concentration increase. The phosphate increase could however have
been too low in order to achieve a complete nitrification.
• The third nitrification test showed the best results from the nitrification tests
assumed to be correlated with an alkalinity increase in combination with an
increase of the phosphate concentrations. Hence, the ammonium-nitrite
transformation was working for all three tests but the second step of
transforming nitrite to nitrate failed. This is most likely due to the phosphate
and alkalinity concentrations, but also to the metal concentrations of the
process water which was not assessed.
• The denitrification test showed partial denitrification which could be related to
a proportion alteration between process water and sludge compared to the
nitrification tests and that denitrification is less sensitive.
The issue prior this study was the low water temperatures in tunnels and in the
biological treatment. All tests in the lab were performed in room temperature and the
nitrification process still failed. Thus, the low water temperature at site is not the
major concern. Other inhibiting parameters on nitrification are metals where the
assessment on the recipients showed increases in several metals that could affect the
nitrification process. The pH could also affect nitrification, though it should be
between 6.5-8.0 which it was for most part of the tests. The proportions in the
solutions of sludge and process water was 500/1 500 ml in the nitrification tests and
900/1 100 ml in denitrification test, which also had the best results. This could be a
factor that affects the nitrification process in combination with increased metal
concentrations, low alkalinity and low phosphate concentrations.
Further studies are required to provide substantial answers and a solution to why the
process water from tunnels is difficult to treat using biological treatment of
nitrification and denitrification. Water is complex and cannot always be analyzed in a
universal perspective, the local conditions differ and a tunnel in other parts of the
world could have other water compositions and surrounding environments. Hence,
further lab analysis on process water with high nitrogen content is crucial which
should focus on phosphate and metal concentration, sludge age and the proportions of
sludge and process water since those parameters were not thoroughly analyzed in this
study.
CHALMERS, Civil and Environmental Engineering, Master’s Thesis 2014:
49 |
Chalmers University of Technology | ii
Concept development of field analysis equipment for mining and
exploration application
Master’s thesis in the Master Degree Programme Product Development
VASUPOL KUNAVUTI
Department of Product and Production Development
Chalmers University of Technology
Abstract
The problems with the exploration process within the mining industry are high investment and
high risks because it takes a very long time and a lot of money to perform the analysis of the
rock sample. With the developed equipment, the process will be shortened, thus benefitting
the mining and exploration companies who are the potential customers.
The aim of this thesis is to create a feasible and production ready concept of equipment for
performing field analysis of rock samples in mining and exploration application.
During the course of this thesis work, a product development process has been used. This
process involves identifying requirements specifications, functional analysis, concept
generation, concept evaluation and concept refinement.
The new product is a newly developed instrument that was improved from the first prototype.
Mainly, the new feature of the machine is that it can scan multiple samples in a standard core
box. The design emphasis of the instrument was on operability, usability, safety,
transportability, environment and compact ability. The aim of the final design was to arrange
the components in order to save space and weight, while maintaining the functionality and
quality of the analysis. The three-dimensional CAD model prototype was created by using a
computer-aided design program. The design was based on the final concept. The special
focuses have been on the new feature of the instrument such as the loading and unloading of
the core box. It is important that the final design is feasible for production, as the aim is to turn
the concept into a fully functioning product. The result of the thesis provides the solutions that
analyses multiple samples in a standard core box. The instrument can withstand the rough and
tough environment, and the exploration process can be performed faster and more accurately.
Keywords: Mining, Exploration, rock analysis, core sample, core box |
Chalmers University of Technology | Chapter 1: Introduction 1
1. Introduction
This chapter introduces the thesis which includes the proposal, objective and scope. In
addition, the current situation of the market within the exploration and mining industry and
the motivation of this thesis are also discussed.
1.1 Background
The mining and exploration industry is growing every year. In Sweden alone, a lot of new
projects are moving towards production. In the upcoming year, the newly established mines
are scheduled to start production and existing mines are expanding their production. For
example, Dannemora iron ore mine north of Stockholm is planning to start production in 2012.
The Northland resource in the north outside Pajala, on the border to Finland, is scheduled to
expand their production to five million ton in 2013. Figure 1.1 (Value of exploration in Sweden
1982-2010) shows that in 2010 investments in exploration rose to 675 million SEK. The number
of exploration permits rose by 40% in the first quarter of 2011 and 192 permits were granted
up until September 2011, while another 50 are being processed at the Mine Inspectorate in
Luleå. 323 Exploration permits have been granted extensions and a further 95 decisions are
pending. (Tomas From, 2011)
Figure 1.1: Value of exploration in Sweden 1982-2010 (million SEK, current price) (Tomas From, 2011)
In order to find the economically feasible mineral resources, the exploration drilling process is
associated with high investment and major risk. Currently, mining and exploration companies
use a portable XRF analyzer and geologist’s judgment to select the core samples from several
drill holes reaching hundreds of meters below the ground and send those samples to
laboratories for accurate analysis. Figure 1.2 illustrates the exploration process. This process
takes a lot of time and is very costly. In view of the information mentioned above, it is a great
opportunity for the company to develop an instrument that can analyze rock core samples
right at the mine and exploration site. With this new instrument, the process will be shortened
since, the instrument can produce the scanned result on-site more accurate and repeatable
than the delivered results by the portable XRF analyzer. The samples no longer need to be
selected because the developed machine can analyze the entire sample from the drill hole.
Exploration Sample Sample Investment
Pre-studies
drilling selection analysis decision
Figure1.2: The exploration process |
Chalmers University of Technology | 2 Chapter 1: Introduction
The company decided to produce a single core scanner to reduce the complexity of the
prototype. The next product will be an improvement on the design and function of the first
prototype and will be produced in the future, based on the final concepts of this thesis.
1.3 Purpose
The aim of this thesis is to generate feasible concepts of X-ray equipment based on the first
prototype for performing field analysis of rock samples in mining and exploration applications.
The developed equipment will be capable of analyzing multiple samples in a standard core
box.
1.4 Objective
The objective is to develop and redesign the current prototype equipment with emphasis on
the following aspects:
Operability, the developed equipment should be able to analyze more rock samples
than the existing prototype.
Usability, the machine should be easy to use according to ergonomics.
Safety, it should be safe to operate.
Transportation, the instrument should be easy to transport.
Environment, the equipment should be able to operate in the rough and tough
environment.
Design, the instrument should be relatively small and light weight.
The intended outcomes of the thesis are the concepts' solutions for the following functions:
loading and unloading the core box, movement of X-ray components and core box, transport
solution, maintenance, environment, components layout, safety and outer design.
1.5 Scope
Within its parameters, the thesis focuses on developing a concept of field analysis equipment.
The thesis does not include programming of PLC or control software, operating software, user
interface software and no physical product. This thesis does not go deep into the subsystem.
However, it does make suggestions on specific components’ layout and requirements. The
developed equipment will be presented with CAD models. The thesis is planned to take 6
months to deliver the expected outcome.
Following areas will be developed during this thesis:
Construction of the list of requirements specification
Generate, evaluate and select the concepts of loading and unloading sample,
movement of samples and components, maintenance and transportation
Designing the layout of the instrument
Designing the cover
Create a CAD model and drawing |
Chalmers University of Technology | 4 Chapter 2: Thesis Process
some part of the report was not updated and needs to be improved. The next step was to
study the sub systems and components of the existing prototype and create component
structures to show the whole system of the machine. The manual of the sub components helps
identify the requirements of the sub system. A process flow model was made to illustrate the
input process and output of the existing prototype.
2.3 Development phase
The Development phase begins with identifying the function and problem. The function
mapping was done to show interference between the functions and to find the most effective
and affected functions. The next step is to generate the concepts. The ideas were established
and sketched to visualize the instrument. A brainstorming session with the company’s
representatives was set up to help develop the idea and to discuss alternative solutions. The
concept table was used to present ideas and solutions of the whole system. The concepts were
narrowed down by using evaluation methods such as Elimination matrices, the evaluations
based on Go/No-Go screening and the Kesselring matrices. The Elimination matrices helped to
eliminate unfeasible ideas. The evaluations based on Go/No-Go screening were used to screen
the concepts. Finally the Kesselring matrices were used to select the promising ideas by
applying rated criteria that were selected from the requirements after the discussion with the
company’s representatives.
The outcome concept is the result from using evaluation methods. It contains the solution of
each function that suited the product. It is however possible that during the product
development process the requirements might have changed or some solution might not be
able to be applied to the equipment. The outcome concepts need to be refined in the next
phase.
2.4 Refinement and deliverables phase
In this phase, the outcome ideas were refined and finalized by assuring that the idea solutions
are feasible and working well together. After establishing the final concept it is important to
create a prototype of the product. The prototype is a visual representation of the concept. A
CAD model was created by using the Autodesk inventor program. It is the best way to design
the machine because if the concept is not feasible, or too complex to be accomplished in
practice, the change of design can be made within the program. Finally, the report has been
finished and presented. |
Chalmers University of Technology | Chapter 3: Theoretical Framework 5
3. Theoretical Framework
The method and tool that were used in the thesis including functional analysis, relations
diagram, concept table, concept evaluation, failure mode and effects analysis and prototyping
will be described in this chapter.
3.1 Functional analysis
Functional analysis is the analysis of the system in terms of its purpose. It involves the
description of the main function and sub-functions of the system. Two functional analysis
methods were applied in the thesis, namely, Process flow model and Function means tree.
3.1.1 Process flow model
The main purpose of a process flow model is to decompose the primary function into sub-
functions. The flows regarding material, energy and information between the sub-functions
are described. The method focuses on input operands, output operands and transforming
function. Figure 3.1 illustrates the process flow model. This method was used in the thesis to
define the flow of the analyzing process of the developed product in order to improve the
desired function and design the new function. (Ullman, 1944)
Input operand Function Output operand
Figure 3.1: Process flow model
3.1.2 Function-means tree
Function mean tree is an approach to capturing the functional requirements and decompose
them in a hierarchical structure. It decomposes the sub-functions and sub-solutions, starting
from the primary function. The purpose of this method is to define the interaction between
the functional domain and physical domains. The tool consists of two types of nodes; functions
- what needs to be done and means - how those can be done. This approach can also be a tool
for idea generation. Figure 3.2 illustrates a function-means tree. Function-means tree was used
in the thesis to provide an understanding of the relationship between functions and the
components. (Ullman, 1944)
Function
Means
Figure 3.2: Function-means tree
3.2 Relations diagram
The relations diagram also known as interrelationship diagram is used to identify cause and
affect relationships between different issues. The relation diagram is created by identifying
factors that are involved in the problem; draw an arrow from the greater influence (the cause) |
Chalmers University of Technology | 6 Chapter 3: Theoretical Framework
to the lesser influence (the one influenced). The factor with the most outgoing arrows will be
the driver or causal factor and the factor with the most incoming arrows will be an outcome or
result. This tool was used in the thesis to organize the functions of the instrument. It identifies
causal factors and help to decide which function to focus on. (Charantimath, 2009)
3.3 Concept table
The use of the concept table (see Table 3.1) was developed for this specific thesis. The concept
table was used for two purposes, to collect the overall solutions in one table and to show the
promising solutions that were evaluated from various methods. The alternative solutions were
indicated by color code. The different colors mean that the solutions were excluded by
different evaluation methods. The total amounts of possible solutions are massive and the
functions of the developed product have unique characteristic, some can affect the complexity
of the overall system, and some have individual properties, which can be influenced from
other functions. To simplify the process, the sub-solutions have been evaluated and selected
from each function before starting to combine the solution. The potential performance of a
solution combination was synthesized to support the decision.
Table 3.1: Concept table
Solution
Function/ 1 2 3 … m
Sub-system
F 1 S1.1 S1.2 S1.3 … S1.m
F 2 S2.1 S2.2 S2.3 … S2.m
F 3 S3.1 S3.2 S3.3 … S3.m
: : : : : :
F n Sm.1 Sn.2 Sn.3 … Sn.m
3.4 Concept Evaluation
Concept evaluation is the process of evaluating and selecting the potential concepts from the
idea generation stage. The aim is to narrow down the amount of concepts and decide which
concepts have the highest possibility for becoming an actual product. The concept evaluation
methods were used in the thesis, including the Elimination matrix, the evaluation based on
Go/No-Go screening and the Kesselring matrix.
3.4.1 Elimination matrix
The Elimination matrix is a filter for the concepts. The evaluation is based on feasibility
judgment. An idea will be excluded if it fails to fulfill one of the following criteria: solving the
main problem, fulfilling the demands of the requirement list, realizable in principle, reasonably
cost effective and safe. Each solution will be judged against the criteria. The result can be: Yes
(+), No (-), Lack of information (?), or Check requirement list (!). And the decision for each
solution can be: Pursue solution (+), Eliminate solution (-), Collect information (?), or Check
requirement list for changes (!). In the thesis, this method was used to eliminate obviously
disqualified solutions. It is a quick step before concept screening and concept scoring. Table
3.2 shows the Elimination matrix. (Pahl, Wallace, & Blessing, 2007) |
Chalmers University of Technology | Chapter 3: Theoretical Framework 7
Table 3.2 Elimination matrix
Selection Criteria
Solution Remarks Decision
A B C D E
S1
S2
:
Sn
3.4.2 Evaluation based on Go/No-Go screening
After that the concepts have passed the Elimination matrix, they will be evaluated against the
criteria defined by the requirements specifications. The questions that generate from the
requirements need to be answered with either Go, or No-Go. This method was modified to
make it more suitable for the thesis. Instead of using the matrix, the evaluation was carried out
in the meeting between the stakeholders. The discussion stemmed from which requirements
should be selected to create the criteria preceding the screening process. (Ullman, 1944)
3.4.3 Kesselring matrix
A Kesselring matrix as shown in Table 3.3 is a concept scoring method. It is used to analyze
potential concepts and finally selecting the final concept. The processes of the Kesselring
approach are:
Choose the criteria based on requirements.
Determine the weight factors.
Assign the grades for fulfillment of the selection criteria.
Calculate the concept scores.
The final concept is selected from the best total score. The Kesselring matrix was applied in the
thesis to evaluate and select the final solution of an individual function. The selection criteria
and weight factors were carefully considered by the stakeholders. Each solution was assigned
a numerical value corresponding to the level of fulfillment of the selection criteria. (Ulrich &
Eppinger, 2008)
Table 3.3 Kesselring matrix
Evaluation Weight Concept 1 Concept 2 Concept m
criteria (W) Value WV Value WV Value WV
1 2 m
(V ) (V ) (V )
1 2 m
A W V W V V W V V W V
1 1A 1 1A 2A 1 2A mA 1 mA
B W V W V V W V V W V
2 1B 2 1B 2B 2 2B mB 2 mB
C W V W V V W V V W V
3 1C 3 1C 2C 3 2C mC 3 mC
: : : : : : : :
n W V W V V W V V W V
n 1n n 1n 2n n 2n mn n mn
∑W ∑ WV ∑ WV ∑ WV
1 2 m
3.5 Failure mode and effects analysis
The Failure mode and effects analysis (FMEA) is a technique for identifying failure potential
and its consequences and the methods to reduce the chance of failure occurring. FMEA is used |
Chalmers University of Technology | 8 Chapter 3: Theoretical Framework
as a design evaluation tool and as a design aide. A failure can be a mechanical failure, based on
any chance that the component, assembly, or system fails to perform its intended function.
Table 3.4 illustrates the FMEA worksheet. It contains four indicators - Potential failure,
Potential effects of failure, Potential cause and Current controls. Also there are three ratings -
Severity, Occurrence and Detection. (Otto & Wood, 2001)
Severity relates to the seriousness of a potential effect of failure on a scale from 1 – 10. 1
implies no effect and 10, implies a very high and hazardous effect. The failure may result from
the product becoming inoperative or unsafe and could result in possible injury. Occurrence is
the probability of failure based on potential cause(s), over the lifespan of the product. This is
indicated on a scale of 1 - 10, where 1 implies failure unlikely and 10 implies failure is almost
inevitable. Detection relates to the possibility that current controls will detect the potential
failure before sending the product to the customer. Finally, a Risk Priority Number (RPN) is
calculated from these values. (Otto & Wood, 2001)
Table 3.4: FMEA Worksheet
Potential Potential Risk priority
Severity Potential Occurrence Current Detection
Function Failure Effects of number
(S) Cause(s) (O) controls (D)
mode Failure (RPN)
3.6 Prototyping
Prototyping is the process of developing an approximation of the product. It provides an
efficient and effective method to modify and improve the product through the testing and
iterative revision. In the thesis, three-dimensional computer modeling (3D CAD) was chosen to
produce the 3D computer aided design models of the actual product. Within the thesis, the 3D
CAD model was used for four purposes. Firstly learning, the prototype is used as a learning
tool. The designed part can be adjusted as many times as it is required until the result is
satisfactory. The estimation can be made to test the outcome and generate discussion for
further development. Secondly communication, the prototype shows the visual of the product.
It not only enhances the communication with the customer and investor, but also the
manufacturing process as well. The three-dimensional CAD model is much easier to
understand than a sketch or drawing. Thirdly integration, the 3D CAD model can demonstrate
the assembly of the product to ensure that the components and sub-systems fit together as
designed. Finally as milestones, if the prototype fulfills the requirements, this can be used as a
yardstick at various stages of production to ensure that each stage of the development of the
product goes well and will perform as expected. (Ulrich & Eppinger, 2008) |
Chalmers University of Technology | Chapter 4: Concept development 9
4. Concept development
This chapter describes the background of the first prototype, identifies the problem and lists
the functions that needed to be improved. The product development process is also outlined
in this chapter starting with establishing the required specifications, functional analysis,
concept generation of the different functions, evaluation and outcome concept.
4.1 The existing prototype
The instrument’s project stemmed from the product development project. The purpose of the
project was “To help the company realize their vision of rationalizing the prospecting process
and to design a concept for a basis for further development.” (Bragsjö, Halonen, Johansson,
Krpo, Sernevi, & Smajic, 2010). The project includes market analysis that was conducted
around four main areas: politics, economics and social issues, technology in prospecting,
possible market segments and competitors and their solutions. The final concept of the project
was called Mobile Automated-Drill Core Analytical Technique (MADCAT) which combined the
concepts of different sub-functions, namely, chassis, user interface, analyzing, core-handling,
environmental protection and positioning of components. (Bragsjö, Halonen, Johansson, Krpo,
Sernevi, & Smajic, 2010)
The company decided to produce a single core scanner instead of a core box scanner to reduce
complexity of the system. The final concept of the development project differs from the
prototype in the following aspects; the number of input samples, sub-system and components
because at that time the development team hadn’t gained the knowledge about the
components and sub systems yet and there existed a lot of unclear information and
uncertainty. Furthermore, the technology was nonexistent or not well known yet. The choices
the development team made for the final concept was the best solution for the situation at
that time.
The existing prototype is a single core scanner. It can only analyze rock samples that are
smaller than 0.5 meters at one time. The instrument uses an X-ray fluorescence (XRF)
technique for the analyzing process. XRF is a non-destructive analytical technique used to
analyze rocks, minerals, sediments and fluids. According to Beckhoff, Kanngießer, Langhoff,
Wedell and Wolff, the XRF principle can be explained as follows. When the inner electron is
excited by photons in the X-ray, the atoms becomes unstable and an outer electron moves
from a higher energy level to fill the gap of the missing inner electron. During the process,
energy is emitted due to the energy difference between the two shells. The released spectrum
reveals a number of characteristic peaks that can be used to detect the identification of the
elements that exist in the sample.
During the study phase, the component structure was created to make an understanding of
the entire components. The main sub- systems of the prototype are power system, motor
system, X-ray system, cooling system, controlling system or PLC, chassis and user interface.
The aim of studying the sub systems and its components is to find an answer to these
questions. “What are the functions of the components?” and “Why does the instrument need
it?” These questions are addressed by studying the manual of each component and then
making a components list by categorizing the components into sub systems and listing the
important requirements and specifications of each sub system. In addition, a list of pros and
cons of each component has been made. Some components might not be suitable for the next |
Chalmers University of Technology | 10 Chapter 4: Concept development
product, such as a high voltage generator with improved technologies, a smaller model with
higher power is available and also a laser sensor that can be useful for the developed product
was found since producing the first prototype. The cooling system of the prototype is huge and
it would probably be better to change it in the developed product of which equipment size is a
crucial requirement. The purpose of the instrument is to analyze rock sample which in the
prototype case is only a single core at a time, but the next machine is aiming to analyze
multiple samples in a standard core box. Figure 4.1 displays the single core scanner.
Figure 4.1: The first prototype
4.2 Requirements specification
The requirements are based on the requirements of the existing prototype and the
requirements of its sub systems and components. The problem is that the requirements of the
existing prototype has not been updated and are unclear because of its lack of detail, hard to
measure and the limit values are not appropriate. The requirements list of the developed
instrument was formulated with additional information and the knowledge gained from the
prototype. Furthermore the requirements of the new function of the instrument were added
into the requirement list.
The big difference between the existing prototype and the new
machine is that instead of a single core scanner, the new one
can scan multiple samples in a standard core box (See Figure
4.2), which means that the developed equipment is a lot more
complex than the existing prototype. It has new functions, such
as loading and unloading the core box and the movement of
the X-ray components. Another issue is the company wants the
new instrument to be compact by putting every component
inside the machine and it must be easy to transport. Hence, the
new design is suitable to be used at the drill site.
For the new product, the new requirements were added and Figure 4.2: The standard core box
some of the target levels were changed in order to support the
new functions and components of the instrument. After discussions with the company, the
first draft of the required specifications was created. The requirements were not complete
because some of the values were unknown and some requirements needed to be discussed
further in the next phase of the development process. For example, the performance of the |
Chalmers University of Technology | Chapter 4: Concept development 11
instrument such as the analytical speed and analysis accuracy depend on the type of material
that the user is looking for. The different settings of the searched material can change the
speed of the analytical process. Also some requirements are vague and hard to interpret.
Another issue that is significant for the instrument is safety with regards to radiation. For this,
both Swedish and EU legislations need to be studied so that it can be implemented effectively.
The list of requirements consists of the type of requirement, either demand or wish, the
requirement statement, target level, validation and evaluation/verification. Table 4.1 shows
the examples requirements that have high priorities for the design criteria. The full required
specifications can be found in appendix B.
Table 4.1: The requirements
Requirement Description
Samples should not move If the samples move during the analyzing process, it will affect
during the analyze process the quality of the result. The user can check the position of the
sample from the photo.
Sound level The instrument should not have a sound level more than 90dB as
this is limited by work place regulation. The sound level can be
measured by using a dB meter.
Operations manual The machine should have a user’s manual to help the user
understand how to set up and execute the processes.
Calibration system To help calibrate the instrument in order to get the best accurate
result. The system should detect the position of the sample.
Internal storage The instrument should have an internal memory used for
recoding the result, position and photo of the sample.
Easy data transfer with A security system that allows only an authorized person to get
authorization lock access to the data.
Ergonomic The working process should consistent with the regulation AFS
1998:1 Belastningsergonomi
Machine reliability The reliability rate should be within 96% of operating time to
fulfill the customer’s expectation.
The analyzing process The sample before and after the analysis should be in the same
should not affect the state.
sample
Machine protection During transportation it is important that the machine is suitably
protected as the sensitive components could get damaged.
Portability It is essential that the machine can be transported to the work
site.
Rock sample (core box) The machine should be able to input a standard core box and
scan the rock samples in both cylinder and half cylinder form,
either small or large iron core.
Environment Protection from the environment in terms of inside temperature,
humidity and water that leak from the core box. Figure 4.3 shows
the environment of drill site. |
Chalmers University of Technology | 14 Chapter 4: Concept development
a) Rotation
Loading and unloading the sample by rotating
the feeding disk.
b) Lift up
The sample cabinet fits under the machine. The core
box is then lifted up to the start position. Unloading
happens on the other side of the machine
c) Video player
Insert the core box horizontally inside the
machine
d) The oven
The feeding hatch opens like an oven door
e) Sledge
As the hatch swings up, the sledge slides out
f) Drawer
The hatch slides out like a drawer
Figure 4.5 a) - f): Loading and unloading idea solutions.
4.4.2 Movement of the X-ray components and core box
The movement of the X-ray components and core box are complicated aspect because the
mechanisms of these functions can affect the size of the instrument. The purpose of the thesis
states clearly that the developed instrument should be compact and as small as possible.
According to the requirements, the size of the machine is limited by the width of a standard
door, 80 cm. The height is 75 cm, being the same as the existing prototype. The length of the
machine depends on the core box size and the mechanism of this function. |
Chalmers University of Technology | Chapter 4: Concept development 15
The movement of the X-ray components and core box should be consistent and work well
together. The scanning area should cover the entire sample in the core box. Three possible
solutions were realized. In later stages, the two of the solutions were eliminated because it
was realized that the core box cannot move along the X-axis. Hence, there was only one option
for this function.
An animation of the movement of the X-ray equipment and core box was made by using
Autodesk inventor for visualization and presented to a big Swedish mining company and
potential customers.
4.4.3 Transport solution
For the transport solution, the first discussion with stakeholders focused on how to deliver the
instrument to the customer safely. Since the machine consists of sensitive components, it is
important to make sure that the instrument before and after transport is in the same state,
and be able to set up and run as fast as possible. In later discussions, it came out that the
possible solution for delivering the instrument from point to point, for example between
drilling site or from the company’s office to working location is to use sub-contractors because
the instrument is of very high value. Sub-contractors will be responsible for any damages to
the machine during transportation. The next topic within transportation is how to protect the
instrument from damage during transportation and how to place the instrument inside the
vehicle.
The transport solution focuses mainly on outer protection and securing the instrument. The
special care must be taken to protect the window and the panel and securing the instrument
in the vehicle. The solution was inspired by the protection process from different ideas such as
the fire truck that uses roll up doors to protect the inside equipment and the container
solution that has special attachments whereby the instrument will be secured. Another
possible solution would be to make forklift pockets for the forklift to be able to move the
instrument.
4.4.4 Maintenance
The maintenance list was created by listing the maintenance requirements of each component
in order to design the layout for further processes. The component manuals are the significant
source of information. The dimensions and technical data, such as operating temperature can
be found in these manuals. The maintenance list consists of the list of components in each
different sub system, the model of component, weight, operating temperature, storage
temperature, humidity, list of maintenance activities and installation notes. The list does not
only consider the components of the first prototype but also includes the new possible
components that might be involved in the new instrument.
The maintenance list gives the information about how often the user needs to access the
components. Some sensitive components need special treatment and more dedicated
attention than others.
Some examples of maintenance would be:
The X-ray tube. In general the glass surface should kept clean and dry at all times, the
cooling circuit inside the tube needs to be inspected and cleaned and the X-ray tube
itself has a specific life span and needs to be changed. |
Chalmers University of Technology | 16 Chapter 4: Concept development
The cooling system requires the checking of the water level and heat exchanger. The
water needs to be refilled and the filter of the pump are checked and cleaned every 3
months.
The maintenance information will affect the components’ layout because the components that
require frequent maintenance and regular checking should be placed in the position that’s
easy to access by the user. Regarding the non-disclosure agreement, the information within
the maintenance list cannot be disclosed because it contains a list of the components, model
and supplier and other sensitive information.
4.4.5 Environment
The main goal of this thesis is to develop the instrument for field use. To achieve that goal the
instrument should withstand the harsh environment of the mining and exploration field. The
factors that can affect the performance of the instrument or jeopardize the operation are
temperature and humidity inside the machine, because the components of each sub-system
can withstand a different range of temperature and humidity. For example, the operating
temperature of the high voltage power supply is between -40 to 85 degree Celsius while the X-
ray tube can operate at 5 to 40 degree Celsius. The temperature inside the instrument should
be adjusted to within a suitable range so that all the systems can operate properly.
The solutions of this problem came from different ideas. Some ideas were based on the
solutions from other machines such as a heater, fan and dehumidifier to control temperature
and humidity or a single component to control both temperature and humidity would be ideal.
Other ideas were to use air curtains or controlled air flow to prevent the outside air from
contaminating air inside the instrument.
4.4.6 Components layout
The idea behind the layout was to fit all components inside the instrument. In the first
prototype, the sub systems were located outside the machine including electronic storage
cabinet, cooling system and High voltage generator. Only the analysis components were placed
inside the machine. In the next product, the functions are more complicated than the
prototype which means it has more components. The challenge of the design was to limit the
size of the instrument. Other factor that might affect the layout was the sensitivity of the sub
systems. For instance, a high voltage generator will create a magnetic field that might affect
other electrical components during the operation.
Initially two component layouts were proposed to the stakeholders. Autodesk inventor was
used to create a visualization of the layout. At this stage, the problems of both layouts were
not yet realized. However, throughout the development process the layout has been
redesigned and improved many times. The details will be discussed in the next chapter,
Refinement.
Another issue regarding components’ layout was the design of the X-ray equipment.
Measurements of the X-ray parts were conducted in order to get the dimensions of the
components.
4.4.7 Safety
The safety of the machine is really important. This aspect was considered to be the first
priority for the design factor. The safety function of the instrument was based on the Swedish |
Chalmers University of Technology | Chapter 4: Concept development 17
Radiation Safety Authority’s rules regarding usage of industrial equipment that contains closed
radiation sources and X-ray tubes. The safety functions included a warning light, alarm system
and security system.
For the warning light, the user should be able to see the light from a 360 degree view around
the machine. The possible solutions were to either put one light on top of the machine or
install four lights on each corner of the machine. The alarm system involves an emergency
stop, status light or alarm sound. Lastly, the security feature was to prevent the user from get
exposed to the X-rays. In other words, the door and hatch of the components that connect to
the analysis cage should be securely locked as to avoid X-rays getting out. The solutions for this
include a key lock, magnetic lock, door sensor and to use a key to start the instrument.
4.4.8 Outer design
The design of the front door that is used for maintenance of the components inside the
analysis chamber such as the X-ray tube was inspire by the doors of a bus, garage door etc.
Examples of the possible solutions are shown in Figure4.6.
a) b) c) d)
ope
Figure 4.6 a) - d): Possible front door solutions
The cover of the instrument was designed by Niklas Arthursson, Product designer. The design
was based on the idea solutions stemming from the idea generation process. The
requirements for the outer design are the instrument should be strong, durable and compact,
but still be visible as high-tech equipment. After discussions with the stakeholders in this
project, the final design of the outer parts was selected. Figure 4.7 shows the final design of
the instrument.
Figure 4.7: Final design of the instrument |
Chalmers University of Technology | 20 Chapter 4: Concept development
positioned lights because it passed the safety criterion which is more serious than the cost
criteria. The alarm warning function was excluded during this process. Even though the
purpose of this function is good, it was not essential for the first development of the
instrument.
4.5.3 Kesselring matrix
The Kesselring matrices, introduced in chapter 3.4.3, were applied after the Go/No-Go
screening. The result from this process will be integrated as the outcome concept. The process
of selection begins with the meeting with the company’s representatives to deal with the
specifications requirements. The problem is that all the requirements need to be prioritized.
By rating each requirement this can help solve the selection process. A rating was given to
each requirement depending on the importance of the requirement.
Hence:
5 = very important
4 = between medium and very important
3 = medium important
2 = between medium and not important
1 = not important
Note that a rating of 1 doesn’t mean that the requirement will not add value to the machine.
The solutions that were evaluated by the Kesselring matrices are loading and unloading the
core box, environment control and the front hatch. Firstly the method of feeding the core box
into the machine, the possible solutions are: video player, the oven, sledge and drawer. When
evaluating each possible mechanism, a rating system was applied to each of the following
criteria: ergonomics/safety, reliability, performance (speed), size, maintenance, environment
(temperature) and cost, the result shows that the video player option has the maximum total
weighted value. This means that the best outcome concept for the feeding function is the
video player solution. The example of the Kesselring matrix: Loading and unloading the core
box is shown in Table 4.4. Secondly the environment control function, the selected criteria
focus on controlling the inside temperature, humidity control and reliability of the instrument.
Finally the front hatch, the main criteria for each function depends on the selected
requirements. Some requirements are significant in all functions because it can affect the
status of the entire system such as reliability and maintenance.
Table 4.4: Kesselring matrix: Loading and unloading the core box
Video player The oven Sledge Drawer
Weight Weighted Weighted Weighted Weighted
Criteria Value Value Value Value
(W) Value Value Value Value
(V1) (V2) (V3) (V4)
(W*V1) (W*V2) (W*V3) (W*V4)
Performance (Speed) 0.11 5 0.53 4 0.42 2 0.21 4 0.42
Weight 0.08 5 0.39 3 0.24 3 0.24 5 0.39
Size 0.11 4 0.42 3 0.32 3 0.32 4 0.42
Ergonomic/Safety 0.13 4 0.53 2 0.26 1 0.13 3 0.39
Reliability 0.13 5 0.66 3 0.39 2 0.26 4 0.53
aesthetic design 0.08 4 0.32 3 0.24 3 0.24 4 0.32
Maintenance 0.11 3 0.32 4 0.42 4 0.42 3 0.32
Control Environment
0.11 4 0.42 2 0.21 2 0.21 4 0.42
(Temp.)
Production 0.05 5 0.26 3 0.16 3 0.16 4 0.21
Cost 0.11 4 0.42 4 0.42 3 0.32 4 0.42
Total 1.00 4.26 3.08 2.50 3.84 |
Chalmers University of Technology | 22 Chapter 4: Concept development
Figure 4.8: Outcome concept
The outer design consists of the front hatch, feeding hatch and the touch screen for human –
machine interaction. The front sides of the instrument are protected by the roll up door. The
edge and the corners of the machine are mounted with bumpers to absorb any impact that
might happen during transportation. With this container concept solution, the instrument can
be placed steadily inside the vehicle. For the maintenance, the tray is a good solution to
prevent dust and dirt from the samples, getting inside the machine. The airflow control and air
filter help collect the dust and purify the air inside the machine. Because the instrument will be
located near drill sites, it is possible that the ambient temperature would be extremely low, for
example -40 degrees as in the case of the mine in Kiruna in the north of Sweden. In order to
ensure that the instrument is operating in this temperature the temperature controller is a
good solution. The heater can increase the temperature inside the instrument until it reaches
the operating temperature. The dehumidifier can be used to adjust the air humidity.
The layout of the machine was designed with the aim to make it as compact as possible, but
still fit all the sub-systems including the X-ray components, electronic storage cabinet, cooling
system, high voltage generator and computer cabinet. The designed layout also
accommodates the movement of the X-ray components and core box.
For the safety aspect, which is the most crucial criteria of instruments using X-ray equipment,
the EU legislation clearly states that portable equipment that contains a closed radiation
should have a well displayed warning light which is lit only when the radiation is present. By
putting the warning lights on the corner of the instrument, it helps fulfill the legislation’s
requirement. If an unexpected event occurs, the instrument has an emergency stop button to
end the process immediately. The most serious issue is the radiation safety. With regards to
safety, the main design goal was to prevent the exposure to the radiation. The user is not
allowed to handle the X-ray components while the equipment is scanning in any circumstance.
The key lock, magnetic lock, door sensor and the use of a key to start the machine not only
solves the radiation safety problem, but also limits access to the control panel, sub-system
cabinet and sensitive components to an authorized person only. The magnetic lock was
installed to secure the front hatch and feeding hatch. The lock responses to the safety PLC |
Chalmers University of Technology | 24 Chapter 5: Refinement
5. Refinement
During the development of the final product, adjustments based on the evaluation methods
always need to be made to improve the concept and to combining the good features of some
alternative solutions that can be adapted for the product. This chapter will describe the
refinement of the requirements specification and the refinement of the final product.
5.1 Refinement of requirements specification
The requirements of the instrument were modified during the development process either
because of the hidden requirements that hadn’t been realized at the beginning of the thesis,
or new knowledge and information that was acquired during the process. These improvements
include clarifying the requirements’ statement, selecting a more accurate target level,
enhancing the method of validation and evaluation, rating the requirement and adding the
new requirements.
The specifications were reformulated for the purpose of justifying the requirements. For
instance, the requirement that starts with, “Easy to …” such as, “Easy to transport to the
mining and exploration site”, “Easy to change between transportation medium” and “Easy to
load/unload the machine into the vehicle” were difficult to define whether it is easy or not.
After the discussion the requirements were changed to “Possible to…”. The changed
requirements fulfill the basic understanding that the instrument should be able to do as stated
in the requirements.
The target level was a tricky aspect because if the target level was too high to achieve, it might
create a problem with the actual production of the machine. The preliminary solution was to
set the target within a required range and the demand and wish target level. After carefully
consideration the target level was changed to a more appropriate and accurate value for
example, the analytical speed is 1mm/s to 10mm/s depending on the type of material, the
analysis accuracy would be 100% accurate for the demand.
The validation and evaluation methods were explored to find the ways to improve the
methods and procedures. For example to validate the mechanical robustness the instrument
should be compared to other machines and equipment that are used in the mining industries.
For the safety function, it is stated in the initial requirement that the instrument should
automatically shut down the X-ray tube when the hatch is open, but when considering the
maintenance aspect, it might occur that the user needs to open some hatch, such as the
computer cabinet or the electronic storage cabinet, while the machine is operating. The final
product considers both the safety and maintenance aspects by designing an inner cage that
prevents radiation leakage. In order to comply with this design, the requirement was added
that the safety sensor be applied only to the hatch that leads to the X-ray tube so the user
should still be able to open the electronic storage and computer hatch while operating.
The main specification change that affected the design is the size of the core box. The original
intention of the design was to have a core box size of 1042 mm x 378 mm x 50 mm, but during
the development, new information came to light that the standard size of the core box varies
depending on the country that uses it. Thus the new dimension of the core box is, 1100 mm x
362 mm x 50 mm making the length of the new core box longer than the first one. |
Chalmers University of Technology | Chapter 5: Refinement 25
5.2 Refinement of the concept
The result from the evaluation methods is only the concept. To make it more feasible the
refinement was needed to fill in the missing detail. Following are the concerns of the
instrument and the solutions to overcome the problems. Some of the concerns were
discovered during the 3D CAD modeling process.
5.2.1 Loading and unloading the core box and feeding hatch
For the method of loading and unloading the core box, the mechanism called “video player”
was chosen. The concerns of this system are the lack of detail about the moving parts, how to
open the hatch and how it works. The modified version of the “video player” integrates the
strength of the “Drawer” concept. The improved concept satisfies all the concerns. The main
features of the mechanism are the movement of the feeding tray and the hatch. The hatch was
designed as a lifting door which could open upwards or downwards. After investigating both
options the conclusion was that the best solution is to open downward because the top of the
hatch could be used to attach the roller or other material to support the tray when loading the
core box. The hatch should have a physical lock as well as the mechanism to prevent it from
falling down. Research was conducted to find an example of a product that used such a
system. The suppliers that can provide a similar type of product were found. The problem of
this solution was realized later. Because of the size of the feeding hatch, it is not only
impossible to make a lifting hatch that has a counter weight mechanism, but also extremely
expensive for a motorized solution. After discussions with the stakeholders, the decision was
made to reduce the complexity of the design by changing the lifting hatch to a normal swing
hatch that opens downward. The decision not only solves the feeding hatch problem but also
simplifies the space problem which is very tight inside the machine.
5.2.2 Cooling system
In the final product, the cooling system is placed in the back of the instrument next to the
electronic storage cabinet. The drawing from the supplier indicates that if the cooling system is
positioned in the back, the air inlet and outlet will be placed on the backside of the equipment
as well. A ventilation problem might occur because the machine will be located close to the
wall where there is restricted air flow. An alternative solution was to place the cooling system
on the side of the machine. The only concern with this was the gap above the cooling system
will be really tight because during the analysis the X-ray components are moving to the left
ends and right ends of the instrument. The problems were solved when a new model of heat
exchanger was discovered, which is very thin but longer. The decision was to put it on the top
of the machine. Even though the total height of the machine will be increased, but when
considering the advantage of having the air inlet and outlet on the top of the machine, this
solution becomes the most suitable choice. The space at the back of the equipment can be
used to install the water tank and pump because it is easier for maintenance when the water
tank is separate from heat exchanger, for instance to refill the water, drain the water out of
the system and cleaning. With this solution, the electrical cabinet can be expanded because
the cooling system cabinet will use less space than in the first layout.
5.2.3 Core box dimension
The most critical issue was the discovery of the new dimensions of the core box. Since the
machine was designed to use the standard core box (1042 mm x 378 mm x 50 mm), the new
box is 1100 mm in length which can jeopardize the design of the machine. The reason behind |
Chalmers University of Technology | Chapter 6: Result 27
6. Result
In this chapter, the final design of the instrument and the limitations of the design will be
described. The topics focus on the feasible and ready to produce concept, the three-
dimensional computer modeling (3D CAD) and the problems that arose during the 3D CAD
modeling.
6.1 Final Design
After the refinement process, it is time to finalize the concept and design the components. The
main parts of the instrument that were designed in this thesis consist of the profile structure,
inner cage, cabinet, analysis equipment, feeding components, front hatch, feeding hatch and
outer plate. The 3D CAD modeling was chosen as a prototyping tool to create the 3D model
and production drawing.
6.1.1 Profile structure
The profile structure is the aluminum frame that supports the inner cage and the outer plate.
The total height of the instrument was expanded because the legs. The reason why the
instrument needs the legs is for the transportation. The machine requires a gap between the
body and the ground at least 130 mm in order to fit the forklift forks. Because the main
function of the frame is to receive the load from the equipment, the beam that is used as the
main structure is thicker than the rest. Triangular profiles were placed at the corners of the
structure for aesthetics and for the installation of the warning lights on the top of the profile.
Figure 6.1 shows a rendering of profile structure created by using the computer-aided design
software, Autodesk Inventor.
Figure 6.1: The profile structure
6.1.2 Inner cage
The inner cage was designed to separate the X-ray beam from the user and other components.
It is important that the primary beam of the X-ray cannot be reached by anyone at any time.
The crucial factor that needs to be considered when designing this part was the radiation
safety. The thickness of the metal sheet required is realized when the outer plate is included,
the total thickness is sufficient to shield against radiation from the X-ray tube. The problem is
when the user opens one of the cabinets during operation the total thickness of the metal
plate is then reduced. The design solution is to use the thicker plate on the cabinet side. In |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.