GitHub

Input output

An input-output system is a system on the form

\[\begin{aligned} M \dot x &= f(x, u, p, t) \\ y &= g(x, u, p, t) \end{aligned}\]

where $x$ is the state, $u$ is the input and $y$ is an output (in some contexts called an observed variables in MTK).

While many uses of ModelingToolkit for simulation do not require the user to think about inputs and outputs (IO), there are certain situations in which handling IO explicitly may be important, such as

  • Linearization
  • Control design
  • System identification
  • FMU export
  • Real-time simulation with external data inputs
  • Custom interfacing with other simulation tools

This documentation page lists utilities that are useful for working with inputs and outputs in ModelingToolkit.

Generating a dynamics function with inputs, $f$

ModelingToolkit can generate the dynamics of a system, the function $M\dot x = f(x, u, p, t)$ above, such that the user can pass not only the state $x$ and parameters $p$ but also an external input $u$. To this end, the function ModelingToolkit.generate_control_function exists.

This function takes a vector of variables that are to be considered inputs, i.e., part of the vector $u$. Alongside returning the function $f$, ModelingToolkit.generate_control_function also returns the chosen state realization of the system after simplification. This vector specifies the order of the state variables $x$, while the user-specified vector u specifies the order of the input variables $u$.

Un-simplified system

This function expects sys to be un-simplified, i.e., mtkcompile or @mtkcompile should not be called on the system before passing it into this function. generate_control_function calls a special version of mtkcompile internally.

Example:

The following example implements a simple first-order system with an input u and state x. The function f is generated using generate_control_function, and the function f is then tested with random input and state values.

using ModelingToolkit
import ModelingToolkit: t_nounits as t, D_nounits as D
@variables x(t)=0 u(t)=0 y(t)
@parameters k = 1
eqs = [D(x) ~ -k * (x + u)
       y ~ x]

@named sys = System(eqs, t)
f, x_sym, ps = ModelingToolkit.generate_control_function(sys, [u], simplify = true);

We can inspect the state realization chosen by MTK

x_sym
1-element Vector{SymbolicUtils.BasicSymbolicImpl.var"typeof(BasicSymbolicImpl)"{SymReal}}:
 x(t)

as expected, x is chosen as the state variable.

Now we can test the generated function f with random input and state values

p = [1]
x = [rand()]
u = [rand()]
@test f[1](x, u, p, 1) ≈ -p[] * (x + u) # Test that the function computes what we expect D(x) = -k*(x + u)
Test Passed

Generating an output function, $g$

ModelingToolkit can also generate a function that computes a specified output of a system, the function $y = g(x, u, p, t)$ above. This is done using the function ModelingToolkit.build_explicit_observed_function. When generating an output function, the user must specify the output variable(s) of interest, as well as any inputs if inputs are relevant to compute the output.

The order of the user-specified output variables determines the order of the output vector $y$.

Input-output variable metadata

See Symbolic Metadata. Metadata specified when creating variables is not directly used by any of the functions above, but the user can use the accessor functions ModelingToolkit.inputs(sys) and ModelingToolkit.outputs(sys) to obtain all variables with such metadata for passing to the functions above. The presence of this metadata is not required for any IO functionality and may be omitted.

Linearization

See Linearization.

Docstrings

    ModelingToolkitBase.generate_control_function — Function
    generate_control_function(sys::ModelingToolkitBase.AbstractSystem, input_ap_name::Union{Symbol, Vector{Symbol}, AnalysisPoint, Vector{AnalysisPoint}}, dist_ap_name::Union{Symbol, Vector{Symbol}, AnalysisPoint, Vector{AnalysisPoint}}; system_modifier = identity, kwargs)

    When called with analysis points as input arguments, we assume that all analysis points corresponds to connections that should be opened (broken). The use case for this is to get rid of input signal blocks, such as Step or Sine, since these are useful for simulation but are not needed when using the plant model in a controller or state estimator.

    source
    (f_oop, f_ip), x_sym, p_sym, io_sys = generate_control_function(
        sys::System,
        inputs                   = unbound_inputs(sys),
        disturbance_inputs       = disturbances(sys);
        known_disturbance_inputs = nothing,
        implicit_dae             = false,
        simplify                 = false,
        split                    = true,
    )

    For a system sys with inputs (as determined by unbound_inputs or user specified), generate functions with additional input argument u

    The returned functions are the out-of-place (f_oop) and in-place (f_ip) forms:

    f_oop : (x,u,p,t)      -> rhs         # basic form
f_oop : (x,u,p,t,w)    -> rhs         # with known_disturbance_inputs
f_ip  : (xout,x,u,p,t) -> nothing     # basic form
f_ip  : (xout,x,u,p,t,w) -> nothing   # with known_disturbance_inputs

    The return values also include the chosen state-realization (the remaining unknowns) x_sym and parameters, in the order they appear as arguments to f.

    Disturbance Handling

    • disturbance_inputs: Unknown disturbance inputs. The generated dynamics will preserve any state and dynamics associated with these disturbances, but the disturbance inputs themselves will not be included as function arguments. This is useful for state observers that estimate unmeasured disturbances.

    • known_disturbance_inputs: Known disturbance inputs. The generated dynamics will preserve state and dynamics, and the disturbance inputs will be added as an additional input argument w to the generated function: (x,u,p,t,w)->rhs.

    Example

    using ModelingToolkitBase: generate_control_function, varmap_to_vars, defaults
f, x_sym, ps = generate_control_function(sys, expression=Val{false}, simplify=false)
p = varmap_to_vars(defaults(sys), ps)
x = varmap_to_vars(defaults(sys), x_sym)
t = 0
f[1](x, inputs, p, t)
    source
    ModelingToolkitBase.build_explicit_observed_function — Function
    build_explicit_observed_function(sys, ts; kwargs...) -> Function(s)

    Generates a function that computes the observed value(s) ts in the system sys, while making the assumption that there are no cycles in the equations.

    Arguments

    • sys: The system for which to generate the function
    • ts: The symbolic observed values whose value should be computed

    Keywords

    • return_inplace = Val(false): If true and the observed value is a vector, then return both the in place and out of place methods. Can take boolean true or false values, but Val(true) or Val(false) is preferred.
    • expression = false: Generates a Julia Exprcomputing the observed value ifexpression` is true
    • eval_expression = false: If true and expression = false, evaluates the returned function in the module eval_module
    • output_type = Array the type of the array generated by a out-of-place vector-valued function
    • param_only = false if true, only allow the generated function to access system parameters
    • inputs = nothing additinoal symbolic variables that should be provided to the generated function
    • disturbance_inputs = nothing symbolic variables representing unknown disturbance inputs (removed from parameters, not added as function arguments)
    • known_disturbance_inputs = nothing symbolic variables representing known disturbance inputs (removed from parameters, added as function arguments)
    • checkbounds = true checks bounds if true when destructuring parameters
    • throw = true if true, throw an error when generating a function for ts that reference variables that do not exist.
    • mkarray: only used if the output is an array (that is, !isscalar(ts) and ts is not a tuple, in which case the result will always be a tuple). Called as mkarray(ts, output_type) where ts are the expressions to put in the array and output_type is the argument of the same name passed to buildexplicitobserved_function.
    • cse = true: Whether to use Common Subexpression Elimination (CSE) to generate a more efficient function.
    • wrap_delays = is_dde(sys): Whether to add an argument for the history function and use it to calculate all delayed variables.

    Returns

    The return value will be either:

    • a single function f_oop if the input is a scalar or if the input is a Vector but return_inplace is false
    • the out of place and in-place functions (f_ip, f_oop) if return_inplace is true and the input is a Vector

    The function(s) f_oop (and potentially f_ip) will be:

    • RuntimeGeneratedFunctions by default,
    • A Julia Expr if expression is true,
    • A directly evaluated Julia function in the module eval_module if eval_expression is true and expression is false.

    The signatures will be of the form g(...) with arguments:

    • output for in-place functions
    • unknowns if param_only is false
    • inputs if inputs is an array of symbolic inputs that should be available in ts
    • p... unconditionally; note that in the case of MTKParameters more than one parameters argument may be present, so it must be splatted
    • t if the system is time-dependent; for example systems of nonlinear equations will not have t
    • known_disturbance_inputs if provided; these are disturbance inputs that are known and provided as arguments

    For example, a function g(op, unknowns, p..., inputs, t, known_disturbances) will be the in-place function generated if return_inplace is true, ts is a vector, an array of inputs inputs is given, known_disturbance_inputs is provided, and param_only is false for a time-dependent system.

    source