Kano001's picture
Upload 3077 files
6a86ad5 verified
raw
history blame
25.5 kB
r"""Activation dynamics for musclotendon models.
Musculotendon models are able to produce active force when they are activated,
which is when a chemical process has taken place within the muscle fibers
causing them to voluntarily contract. Biologically this chemical process (the
diffusion of :math:`\textrm{Ca}^{2+}` ions) is not the input in the system,
electrical signals from the nervous system are. These are termed excitations.
Activation dynamics, which relates the normalized excitation level to the
normalized activation level, can be modeled by the models present in this
module.
"""
from abc import ABC, abstractmethod
from functools import cached_property
from sympy.core.symbol import Symbol
from sympy.core.numbers import Float, Integer, Rational
from sympy.functions.elementary.hyperbolic import tanh
from sympy.matrices.dense import MutableDenseMatrix as Matrix, zeros
from sympy.physics.biomechanics._mixin import _NamedMixin
from sympy.physics.mechanics import dynamicsymbols
__all__ = [
'ActivationBase',
'FirstOrderActivationDeGroote2016',
'ZerothOrderActivation',
]
class ActivationBase(ABC, _NamedMixin):
"""Abstract base class for all activation dynamics classes to inherit from.
Notes
=====
Instances of this class cannot be directly instantiated by users. However,
it can be used to created custom activation dynamics types through
subclassing.
"""
def __init__(self, name):
"""Initializer for ``ActivationBase``."""
self.name = str(name)
# Symbols
self._e = dynamicsymbols(f"e_{name}")
self._a = dynamicsymbols(f"a_{name}")
@classmethod
@abstractmethod
def with_defaults(cls, name):
"""Alternate constructor that provides recommended defaults for
constants."""
pass
@property
def excitation(self):
"""Dynamic symbol representing excitation.
Explanation
===========
The alias ``e`` can also be used to access the same attribute.
"""
return self._e
@property
def e(self):
"""Dynamic symbol representing excitation.
Explanation
===========
The alias ``excitation`` can also be used to access the same attribute.
"""
return self._e
@property
def activation(self):
"""Dynamic symbol representing activation.
Explanation
===========
The alias ``a`` can also be used to access the same attribute.
"""
return self._a
@property
def a(self):
"""Dynamic symbol representing activation.
Explanation
===========
The alias ``activation`` can also be used to access the same attribute.
"""
return self._a
@property
@abstractmethod
def order(self):
"""Order of the (differential) equation governing activation."""
pass
@property
@abstractmethod
def state_vars(self):
"""Ordered column matrix of functions of time that represent the state
variables.
Explanation
===========
The alias ``x`` can also be used to access the same attribute.
"""
pass
@property
@abstractmethod
def x(self):
"""Ordered column matrix of functions of time that represent the state
variables.
Explanation
===========
The alias ``state_vars`` can also be used to access the same attribute.
"""
pass
@property
@abstractmethod
def input_vars(self):
"""Ordered column matrix of functions of time that represent the input
variables.
Explanation
===========
The alias ``r`` can also be used to access the same attribute.
"""
pass
@property
@abstractmethod
def r(self):
"""Ordered column matrix of functions of time that represent the input
variables.
Explanation
===========
The alias ``input_vars`` can also be used to access the same attribute.
"""
pass
@property
@abstractmethod
def constants(self):
"""Ordered column matrix of non-time varying symbols present in ``M``
and ``F``.
Only symbolic constants are returned. If a numeric type (e.g. ``Float``)
has been used instead of ``Symbol`` for a constant then that attribute
will not be included in the matrix returned by this property. This is
because the primary use of this property attribute is to provide an
ordered sequence of the still-free symbols that require numeric values
during code generation.
Explanation
===========
The alias ``p`` can also be used to access the same attribute.
"""
pass
@property
@abstractmethod
def p(self):
"""Ordered column matrix of non-time varying symbols present in ``M``
and ``F``.
Only symbolic constants are returned. If a numeric type (e.g. ``Float``)
has been used instead of ``Symbol`` for a constant then that attribute
will not be included in the matrix returned by this property. This is
because the primary use of this property attribute is to provide an
ordered sequence of the still-free symbols that require numeric values
during code generation.
Explanation
===========
The alias ``constants`` can also be used to access the same attribute.
"""
pass
@property
@abstractmethod
def M(self):
"""Ordered square matrix of coefficients on the LHS of ``M x' = F``.
Explanation
===========
The square matrix that forms part of the LHS of the linear system of
ordinary differential equations governing the activation dynamics:
``M(x, r, t, p) x' = F(x, r, t, p)``.
"""
pass
@property
@abstractmethod
def F(self):
"""Ordered column matrix of equations on the RHS of ``M x' = F``.
Explanation
===========
The column matrix that forms the RHS of the linear system of ordinary
differential equations governing the activation dynamics:
``M(x, r, t, p) x' = F(x, r, t, p)``.
"""
pass
@abstractmethod
def rhs(self):
"""
Explanation
===========
The solution to the linear system of ordinary differential equations
governing the activation dynamics:
``M(x, r, t, p) x' = F(x, r, t, p)``.
"""
pass
def __eq__(self, other):
"""Equality check for activation dynamics."""
if type(self) != type(other):
return False
if self.name != other.name:
return False
return True
def __repr__(self):
"""Default representation of activation dynamics."""
return f'{self.__class__.__name__}({self.name!r})'
class ZerothOrderActivation(ActivationBase):
"""Simple zeroth-order activation dynamics mapping excitation to
activation.
Explanation
===========
Zeroth-order activation dynamics are useful in instances where you want to
reduce the complexity of your musculotendon dynamics as they simple map
exictation to activation. As a result, no additional state equations are
introduced to your system. They also remove a potential source of delay
between the input and dynamics of your system as no (ordinary) differential
equations are involed.
"""
def __init__(self, name):
"""Initializer for ``ZerothOrderActivation``.
Parameters
==========
name : str
The name identifier associated with the instance. Must be a string
of length at least 1.
"""
super().__init__(name)
# Zeroth-order activation dynamics has activation equal excitation so
# overwrite the symbol for activation with the excitation symbol.
self._a = self._e
@classmethod
def with_defaults(cls, name):
"""Alternate constructor that provides recommended defaults for
constants.
Explanation
===========
As this concrete class doesn't implement any constants associated with
its dynamics, this ``classmethod`` simply creates a standard instance
of ``ZerothOrderActivation``. An implementation is provided to ensure
a consistent interface between all ``ActivationBase`` concrete classes.
"""
return cls(name)
@property
def order(self):
"""Order of the (differential) equation governing activation."""
return 0
@property
def state_vars(self):
"""Ordered column matrix of functions of time that represent the state
variables.
Explanation
===========
As zeroth-order activation dynamics simply maps excitation to
activation, this class has no associated state variables and so this
property return an empty column ``Matrix`` with shape (0, 1).
The alias ``x`` can also be used to access the same attribute.
"""
return zeros(0, 1)
@property
def x(self):
"""Ordered column matrix of functions of time that represent the state
variables.
Explanation
===========
As zeroth-order activation dynamics simply maps excitation to
activation, this class has no associated state variables and so this
property return an empty column ``Matrix`` with shape (0, 1).
The alias ``state_vars`` can also be used to access the same attribute.
"""
return zeros(0, 1)
@property
def input_vars(self):
"""Ordered column matrix of functions of time that represent the input
variables.
Explanation
===========
Excitation is the only input in zeroth-order activation dynamics and so
this property returns a column ``Matrix`` with one entry, ``e``, and
shape (1, 1).
The alias ``r`` can also be used to access the same attribute.
"""
return Matrix([self._e])
@property
def r(self):
"""Ordered column matrix of functions of time that represent the input
variables.
Explanation
===========
Excitation is the only input in zeroth-order activation dynamics and so
this property returns a column ``Matrix`` with one entry, ``e``, and
shape (1, 1).
The alias ``input_vars`` can also be used to access the same attribute.
"""
return Matrix([self._e])
@property
def constants(self):
"""Ordered column matrix of non-time varying symbols present in ``M``
and ``F``.
Only symbolic constants are returned. If a numeric type (e.g. ``Float``)
has been used instead of ``Symbol`` for a constant then that attribute
will not be included in the matrix returned by this property. This is
because the primary use of this property attribute is to provide an
ordered sequence of the still-free symbols that require numeric values
during code generation.
Explanation
===========
As zeroth-order activation dynamics simply maps excitation to
activation, this class has no associated constants and so this property
return an empty column ``Matrix`` with shape (0, 1).
The alias ``p`` can also be used to access the same attribute.
"""
return zeros(0, 1)
@property
def p(self):
"""Ordered column matrix of non-time varying symbols present in ``M``
and ``F``.
Only symbolic constants are returned. If a numeric type (e.g. ``Float``)
has been used instead of ``Symbol`` for a constant then that attribute
will not be included in the matrix returned by this property. This is
because the primary use of this property attribute is to provide an
ordered sequence of the still-free symbols that require numeric values
during code generation.
Explanation
===========
As zeroth-order activation dynamics simply maps excitation to
activation, this class has no associated constants and so this property
return an empty column ``Matrix`` with shape (0, 1).
The alias ``constants`` can also be used to access the same attribute.
"""
return zeros(0, 1)
@property
def M(self):
"""Ordered square matrix of coefficients on the LHS of ``M x' = F``.
Explanation
===========
The square matrix that forms part of the LHS of the linear system of
ordinary differential equations governing the activation dynamics:
``M(x, r, t, p) x' = F(x, r, t, p)``.
As zeroth-order activation dynamics have no state variables, this
linear system has dimension 0 and therefore ``M`` is an empty square
``Matrix`` with shape (0, 0).
"""
return Matrix([])
@property
def F(self):
"""Ordered column matrix of equations on the RHS of ``M x' = F``.
Explanation
===========
The column matrix that forms the RHS of the linear system of ordinary
differential equations governing the activation dynamics:
``M(x, r, t, p) x' = F(x, r, t, p)``.
As zeroth-order activation dynamics have no state variables, this
linear system has dimension 0 and therefore ``F`` is an empty column
``Matrix`` with shape (0, 1).
"""
return zeros(0, 1)
def rhs(self):
"""Ordered column matrix of equations for the solution of ``M x' = F``.
Explanation
===========
The solution to the linear system of ordinary differential equations
governing the activation dynamics:
``M(x, r, t, p) x' = F(x, r, t, p)``.
As zeroth-order activation dynamics have no state variables, this
linear has dimension 0 and therefore this method returns an empty
column ``Matrix`` with shape (0, 1).
"""
return zeros(0, 1)
class FirstOrderActivationDeGroote2016(ActivationBase):
r"""First-order activation dynamics based on De Groote et al., 2016 [1]_.
Explanation
===========
Gives the first-order activation dynamics equation for the rate of change
of activation with respect to time as a function of excitation and
activation.
The function is defined by the equation:
.. math::
\frac{da}{dt} = \left(\frac{\frac{1}{2} + a0}{\tau_a \left(\frac{1}{2}
+ \frac{3a}{2}\right)} + \frac{\left(\frac{1}{2}
+ \frac{3a}{2}\right) \left(\frac{1}{2} - a0\right)}{\tau_d}\right)
\left(e - a\right)
where
.. math::
a0 = \frac{\tanh{\left(b \left(e - a\right) \right)}}{2}
with constant values of :math:`tau_a = 0.015`, :math:`tau_d = 0.060`, and
:math:`b = 10`.
References
==========
.. [1] De Groote, F., Kinney, A. L., Rao, A. V., & Fregly, B. J., Evaluation
of direct collocation optimal control problem formulations for
solving the muscle redundancy problem, Annals of biomedical
engineering, 44(10), (2016) pp. 2922-2936
"""
def __init__(self,
name,
activation_time_constant=None,
deactivation_time_constant=None,
smoothing_rate=None,
):
"""Initializer for ``FirstOrderActivationDeGroote2016``.
Parameters
==========
activation time constant : Symbol | Number | None
The value of the activation time constant governing the delay
between excitation and activation when excitation exceeds
activation.
deactivation time constant : Symbol | Number | None
The value of the deactivation time constant governing the delay
between excitation and activation when activation exceeds
excitation.
smoothing_rate : Symbol | Number | None
The slope of the hyperbolic tangent function used to smooth between
the switching of the equations where excitation exceed activation
and where activation exceeds excitation. The recommended value to
use is ``10``, but values between ``0.1`` and ``100`` can be used.
"""
super().__init__(name)
# Symbols
self.activation_time_constant = activation_time_constant
self.deactivation_time_constant = deactivation_time_constant
self.smoothing_rate = smoothing_rate
@classmethod
def with_defaults(cls, name):
r"""Alternate constructor that will use the published constants.
Explanation
===========
Returns an instance of ``FirstOrderActivationDeGroote2016`` using the
three constant values specified in the original publication.
These have the values:
:math:`tau_a = 0.015`
:math:`tau_d = 0.060`
:math:`b = 10`
"""
tau_a = Float('0.015')
tau_d = Float('0.060')
b = Float('10.0')
return cls(name, tau_a, tau_d, b)
@property
def activation_time_constant(self):
"""Delay constant for activation.
Explanation
===========
The alias ```tau_a`` can also be used to access the same attribute.
"""
return self._tau_a
@activation_time_constant.setter
def activation_time_constant(self, tau_a):
if hasattr(self, '_tau_a'):
msg = (
f'Can\'t set attribute `activation_time_constant` to '
f'{repr(tau_a)} as it is immutable and already has value '
f'{self._tau_a}.'
)
raise AttributeError(msg)
self._tau_a = Symbol(f'tau_a_{self.name}') if tau_a is None else tau_a
@property
def tau_a(self):
"""Delay constant for activation.
Explanation
===========
The alias ``activation_time_constant`` can also be used to access the
same attribute.
"""
return self._tau_a
@property
def deactivation_time_constant(self):
"""Delay constant for deactivation.
Explanation
===========
The alias ``tau_d`` can also be used to access the same attribute.
"""
return self._tau_d
@deactivation_time_constant.setter
def deactivation_time_constant(self, tau_d):
if hasattr(self, '_tau_d'):
msg = (
f'Can\'t set attribute `deactivation_time_constant` to '
f'{repr(tau_d)} as it is immutable and already has value '
f'{self._tau_d}.'
)
raise AttributeError(msg)
self._tau_d = Symbol(f'tau_d_{self.name}') if tau_d is None else tau_d
@property
def tau_d(self):
"""Delay constant for deactivation.
Explanation
===========
The alias ``deactivation_time_constant`` can also be used to access the
same attribute.
"""
return self._tau_d
@property
def smoothing_rate(self):
"""Smoothing constant for the hyperbolic tangent term.
Explanation
===========
The alias ``b`` can also be used to access the same attribute.
"""
return self._b
@smoothing_rate.setter
def smoothing_rate(self, b):
if hasattr(self, '_b'):
msg = (
f'Can\'t set attribute `smoothing_rate` to {b!r} as it is '
f'immutable and already has value {self._b!r}.'
)
raise AttributeError(msg)
self._b = Symbol(f'b_{self.name}') if b is None else b
@property
def b(self):
"""Smoothing constant for the hyperbolic tangent term.
Explanation
===========
The alias ``smoothing_rate`` can also be used to access the same
attribute.
"""
return self._b
@property
def order(self):
"""Order of the (differential) equation governing activation."""
return 1
@property
def state_vars(self):
"""Ordered column matrix of functions of time that represent the state
variables.
Explanation
===========
The alias ``x`` can also be used to access the same attribute.
"""
return Matrix([self._a])
@property
def x(self):
"""Ordered column matrix of functions of time that represent the state
variables.
Explanation
===========
The alias ``state_vars`` can also be used to access the same attribute.
"""
return Matrix([self._a])
@property
def input_vars(self):
"""Ordered column matrix of functions of time that represent the input
variables.
Explanation
===========
The alias ``r`` can also be used to access the same attribute.
"""
return Matrix([self._e])
@property
def r(self):
"""Ordered column matrix of functions of time that represent the input
variables.
Explanation
===========
The alias ``input_vars`` can also be used to access the same attribute.
"""
return Matrix([self._e])
@property
def constants(self):
"""Ordered column matrix of non-time varying symbols present in ``M``
and ``F``.
Only symbolic constants are returned. If a numeric type (e.g. ``Float``)
has been used instead of ``Symbol`` for a constant then that attribute
will not be included in the matrix returned by this property. This is
because the primary use of this property attribute is to provide an
ordered sequence of the still-free symbols that require numeric values
during code generation.
Explanation
===========
The alias ``p`` can also be used to access the same attribute.
"""
constants = [self._tau_a, self._tau_d, self._b]
symbolic_constants = [c for c in constants if not c.is_number]
return Matrix(symbolic_constants) if symbolic_constants else zeros(0, 1)
@property
def p(self):
"""Ordered column matrix of non-time varying symbols present in ``M``
and ``F``.
Explanation
===========
Only symbolic constants are returned. If a numeric type (e.g. ``Float``)
has been used instead of ``Symbol`` for a constant then that attribute
will not be included in the matrix returned by this property. This is
because the primary use of this property attribute is to provide an
ordered sequence of the still-free symbols that require numeric values
during code generation.
The alias ``constants`` can also be used to access the same attribute.
"""
constants = [self._tau_a, self._tau_d, self._b]
symbolic_constants = [c for c in constants if not c.is_number]
return Matrix(symbolic_constants) if symbolic_constants else zeros(0, 1)
@property
def M(self):
"""Ordered square matrix of coefficients on the LHS of ``M x' = F``.
Explanation
===========
The square matrix that forms part of the LHS of the linear system of
ordinary differential equations governing the activation dynamics:
``M(x, r, t, p) x' = F(x, r, t, p)``.
"""
return Matrix([Integer(1)])
@property
def F(self):
"""Ordered column matrix of equations on the RHS of ``M x' = F``.
Explanation
===========
The column matrix that forms the RHS of the linear system of ordinary
differential equations governing the activation dynamics:
``M(x, r, t, p) x' = F(x, r, t, p)``.
"""
return Matrix([self._da_eqn])
def rhs(self):
"""Ordered column matrix of equations for the solution of ``M x' = F``.
Explanation
===========
The solution to the linear system of ordinary differential equations
governing the activation dynamics:
``M(x, r, t, p) x' = F(x, r, t, p)``.
"""
return Matrix([self._da_eqn])
@cached_property
def _da_eqn(self):
HALF = Rational(1, 2)
a0 = HALF * tanh(self._b * (self._e - self._a))
a1 = (HALF + Rational(3, 2) * self._a)
a2 = (HALF + a0) / (self._tau_a * a1)
a3 = a1 * (HALF - a0) / self._tau_d
activation_dynamics_equation = (a2 + a3) * (self._e - self._a)
return activation_dynamics_equation
def __eq__(self, other):
"""Equality check for ``FirstOrderActivationDeGroote2016``."""
if type(self) != type(other):
return False
self_attrs = (self.name, self.tau_a, self.tau_d, self.b)
other_attrs = (other.name, other.tau_a, other.tau_d, other.b)
if self_attrs == other_attrs:
return True
return False
def __repr__(self):
"""Representation of ``FirstOrderActivationDeGroote2016``."""
return (
f'{self.__class__.__name__}({self.name!r}, '
f'activation_time_constant={self.tau_a!r}, '
f'deactivation_time_constant={self.tau_d!r}, '
f'smoothing_rate={self.b!r})'
)