component-model 0.0.1

Constructs a Functional Mockup Interface component model from a python script (fulfilling some requirements)

The package extends the PythonFMU package. It includes the necessary modules to construct a component model according to the fmi, OSP and DNV-RP-0513 standards with focus on the following features:

• seamless translation of a Python model to an FMU package with minimal overhead (definition of FMU interface)

• support of vector variables (numpy)

• support of variable units and display units

• support of range checking of variables

Features which facilitate Assurance of Simulation Models, DNV-RP-0513 shall have a special focus in this package.

Getting Started

A new model can consist of any python code. To turn the python code into an FMU the following is necessary

1. The model code is wrapped into a Python class which inherits from Model

2. The exposed interface variables (model parameters, input- and output connectors) are defined as Variable objects

3. The (model).do_step( time, dt) function of the model class is extended with model internal code, i.e. model evolves from time to time+dt.

4. Calling the method Model.build() will then compile the FMU and package it into a suitable FMU file.

See the files example_models/bouncing_ball.py and tests/test_make_bouncingBall.py supplied with this package as a simple example of this process. The first file defines the model class and the second file demonstrates the process of making the FMU and using it within fmpy and OSP.

1. Install the component_model package: pip install component_model

2. Software dependencies: PythonFMU, numpy, pint, uuid, ElementTree

3. Latest releases: Version 0.1, based on PythonFMU 0.64

Usage example

This is another BouncingBall example, using 3D vectors and units.

from math import sqrt

import numpy as np

from component_model.model import Model
from component_model.variable import Variable


class BouncingBall(Model):
    """Another BouncingBall model, made in Python and using Model and Variable to construct a FMU.

    Special features:

    * The ball has a 3-D vector as position and speed
    * As output variable the model estimates the next bouncing point
    * As input variables, the restitution coefficient `e` and the ground angle at the bouncing point can be changed.
    * Internal units are SI (m,s,rad)

    Args:
        pos (np.array)=(0,0,1): The 3-D position in of the ball at time [m]
        speed (np.array)=(1,0,0): The 3-D speed of the ball at time [m/s]
        g (float)=9.81: The gravitational acceleration [m/s^2]
        e (float)=0.9: The coefficient of restitution (dimensionless): |speed after| / |speed before| collision
        min_speed_z (float)=1e-6: The minimum speed in z-direction when bouncing stops [m/s]
    """

    def __init__(
        self,
        name: str = "BouncingBall_3D",
        description="Another BouncingBall model, made in Python and using Model and Variable to construct a FMU",
        pos: tuple = (0, 0, 10),
        speed: tuple = (1, 0, 0),
        g: float = 9.81,
        e: float = 0.9,
        min_speed_z: float = 1e-6,
        **kwargs,
    ):
        super().__init__(name, description, **kwargs)
        self.pos = np.array(pos, dtype=float)
        self.speed = np.array(speed, dtype=float)
        self.a = np.array((0, 0, -g), float)
        self.g = g
        self.e = e
        self.min_speed_z = min_speed_z
        self.stopped = False
        self.time = 0.0
        self.t_bounce, self.p_bounce = self.next_bounce()
        self._interface_variables()

    def _interface_variables(self):
        """Define the FMU2 interface variables, using the variable interface."""
        self._pos = Variable(
            self,
            name="pos",
            description="The 3D position of the ball [m] (height in inch as displayUnit example.",
            causality="output",
            variability="continuous",
            initial="exact",
            start=(str(self.pos[0]) + "m", str(self.pos[1]) + "m", str(self.pos[2]) + "inch"),
            rng=((0, "100 m"), None, (0, "10 m")),
        )
        self._speed = Variable(
            self,
            name="speed",
            description="The 3D speed of the ball, i.e. d pos / dt [m/s]",
            causality="output",
            variability="continuous",
            initial="exact",
            start=tuple(str(x) + "m/s" for x in self.speed),
            rng=((0, "1 m/s"), None, ("-100 m/s", "100 m/s")),
        )
        self._g = Variable(
            self,
            name="g",
            description="The gravitational acceleration (absolute value).",
            causality="parameter",
            variability="fixed",
            start=str(self.g) + "m/s^2",
            rng=(),
        )
        self._e = Variable(
            self,
            name="e",
            description="The coefficient of restitution, i.e. |speed after| / |speed before| bounce.",
            causality="parameter",
            variability="fixed",
            start=self.e,
            rng=(),
        )
        self._p_bounce = Variable(
            self,
            name="p_bounce",
            description="The expected position of the next bounce as 3D vector",
            causality="output",
            variability="continuous",
            start=tuple(str(x) for x in self.p_bounce),
            rng=(),
        )

    def do_step(self, time, dt):
        """Perform a simulation step from `time` to `time + dt`."""
        if not super().do_step(time, dt):
            return False
        self.t_bounce, self.p_bounce = self.next_bounce()
        while dt > self.t_bounce:  # if the time is this long
            dt -= self.t_bounce
            self.pos = self.p_bounce
            self.speed -= self.a * self.t_bounce  # speed before bouncing
            self.speed[2] = -self.speed[2]  # speed after bouncing if e==1.0
            self.speed *= self.e  # speed reduction due to coefficient of restitution
            if self.speed[2] < self.min_speed_z:
                self.stopped = True
                self.a[2] = 0.0
                self.speed[2] = 0.0
                self.pos[2] = 0.0
            self.t_bounce, self.p_bounce = self.next_bounce()
        self.speed += self.a * dt
        self.pos += self.speed * dt + 0.5 * self.a * dt**2
        if self.pos[2] < 0:
            self.pos[2] = 0
        # print(f"@{time}. pos {self.pos}, speed {self.speed}, bounce {self.t_bounce}")
        return True

    def next_bounce(self):
        """Calculate time until next bounce and position where the ground will be hit,
        based on current time, pos and speed.
        """
        if self.stopped:  # stopped bouncing
            return (1e300, np.array((1e300, 1e300, 0), float))
            # return ( float('inf'), np.array( (float('inf'), float('inf'), 0), float))
        else:
            t_bounce = (self.speed[2] + sqrt(self.speed[2] ** 2 + 2 * self.g * self.pos[2])) / self.g
            p_bounce = self.pos + self.speed * t_bounce  # linear. not correct for z-direction!
            p_bounce[2] = 0
            return (t_bounce, p_bounce)

    def setup_experiment(self, start: float):
        """Set initial (non-interface) variables."""
        super().setup_experiment(start)
        self.stopped = False

The following might be noted:

• The interface variables are defined in a separate local method _interface_variables, keeping it separate from the model code.

• The do_step() method contains the essential code, describing how the ball moves through the air. It calls the super().do_step() method, which is essential to link it to Model. The return True statement is also essential for the working of the emerging FMU.

• The next_bounce() method is a helper method.

• In addition to the extension of do_step(), here also the setup_experiment() method is extended. Local (non-interface) variables can thus be initialized in a convenient way.

It should be self-evident that thorough testing of any model is necessary before translation to a FMU. The simulation orchestration engine (e.g. OSP) used to run FMUs obfuscates error messages, such that first stage assurance of a model should aways done using e.g. pytest.

The minimal code to make the FMU file package is

from component_model.model import Model
from fmpy.util import fmu_info

asBuilt = Model.build("../component_model/example_models/bouncing_ball.py")
info = fmu_info(asBuilt.name)  # not necessary, but it lists essential properties of the FMU

The model can then be run using fmpy

from fmpy import plot_result, simulate_fmu

result = simulate_fmu(
    "BouncingBall.fmu",
    stop_time=3.0,
    step_size=0.1,
    validate=True,
    solver="Euler",
    debug_logging=True,
    logger=print,
    start_values={"pos[2]": 2}, # optional start value settings
)
plot_result(result)

Similarly, the model can be run using OSP (or rather libcosimpy - OSP wrapped into Python):

from libcosimpy.CosimEnums import CosimExecutionState
from libcosimpy.CosimExecution import CosimExecution
from libcosimpy.CosimSlave import CosimLocalSlave

sim = CosimExecution.from_step_size(step_size=1e7)  # empty execution object with fixed time step in nanos
bb = CosimLocalSlave(fmu_path="./BouncingBall.fmu", instance_name="bb")

print("SLAVE", bb, sim.status())

ibb = sim.add_local_slave(bb)
assert ibb == 0, f"local slave number {ibb}"

reference_dict = {var_ref.name.decode(): var_ref.reference for var_ref in sim.slave_variables(ibb)}

# Set initial values
sim.real_initial_value(ibb, reference_dict["pos[2]"], 2.0)

sim_status = sim.status()
assert sim_status.current_time == 0
assert CosimExecutionState(sim_status.state) == CosimExecutionState.STOPPED
infos = sim.slave_infos()
print("INFOS", infos)

# Simulate for 1 second
sim.simulate_until(target_time=3e9)

This is admittedly more complex than the fmpy example, but it should be emphasised that fmpy is made for single component model simulation (testing), while OSP is made for multi-component systems.

Contribute

Anybody in the FMU and OSP community is welcome to contribute to this code, to make it better, and especially including other features from model assurance, as we firmly believe that trust in our models is needed if we want to base critical decisions on the support from these models.