Evolution of the wear state variables in the PneumaticValveWithWear model

[matthieucx]

Hello! I am playing around a little bit, discovering the library and looked into the code for the valve with wear model.

From what I understand, in the Valve model, the wear rate is a parameter. It's a property of the system that does not evolve. In the ValveWithWear, the wear rate is dependent on time and evolves, as part of the state of the machine. Is that correct ?

If yes, looking at the next_state() implementation of the ValveWithWear, I am not too sure of how the wear evolves.

def next_state(self, x, u, dt: float):
    with warnings.catch_warnings():
        warnings.simplefilter("ignore")
        self.parameters['wb'] = x['wb']
        self.parameters['wi'] = x['wi']
        self.parameters['wk'] = x['wk']
        self.parameters['wr'] = x['wr']
        self.parameters['wt'] = x['wt']
    next_x = PneumaticValveBase.next_state(self, x, u, dt)

    # Append this way because the keys in the structure but the values are
    # missing - this is due to the behavior of subclassed models calling
    # their parent functions.

    next_x.matrix = np.vstack((next_x.matrix, np.array([
        np.atleast_1d(x['wb']),
        np.atleast_1d(x['wi']),
        np.atleast_1d(x['wk']),
        np.atleast_1d(x['wr']),
        np.atleast_1d(x['wt'])
    ])))
    return next_x

So here we update the wear parameters to be the same as those in the current state. This is because the params are used in the next state calculation (as we use the base model function, in which the wear rate is a parameter and not part of the state). Relevant code below.

Then, we compute the next state using the base model function.
Then we append the wear parameters to the state matrix. We append the values of the current state (which is logical, since, again, the state returned by the base class' function does not contain the wear values).

But then, here we have x+1[w*] = x[w*] ?

Full next_state() of the base model

def next_state(self, x, u, dt: float):
  params = self.parameters  # optimization
  # ...
  kdot = -params['wk']*abs(x['v']*springForce)
  rdot = params['wr']*abs(x['v']*friction)
  Aidot = params['wi']*abs(x['v']*friction)
  # ...
  return self.StateContainer(np.array([
      np.atleast_1d(x['Aeb'] + params['wb'] * dt),     # Aeb
      np.atleast_1d(x['Aet'] + params['wt'] * dt),     # Aet
      np.atleast_1d(x['Ai'] + Aidot * dt),             # Ai
      # ...
]))

Thanks in advance for any help on understanding what is happening here. Sorry if I missed something obvious!

[teubert]

Hello @chefmtt,

Thanks for your question. You're right that the wear is added as a state so it can change freely. However,t doesn't (as the model is now) change within a simulation. In next_state, the wear parameters are added back in, unchanged.

This might seem odd for someone learning prog_models, as it will behave functionally identically as having the wear rates as parameters. Where this makes a difference is when combined with prog_algs. Some users are trying to estimate the wear parameters online using a StateEstimator (see https://nasa.github.io/progpy/prog_algs_guide.html#state-estimation). For these users, having the wear parameters in the state allows the state estimator to estimate those wear rates with the other states.

This is more of an edge case. Most users estimate the wear parameters using parameter estimation and then simulate forward using the base PneumaticValve model. Some users may even update those parameters occasionally by re-running estimate_params. But the *WithWear models are included for the subset of users who would like to update the wear rates online with a state estimator.

Does that answer your question? Let me know if you have any additional questions.

[matthieucx]

Hello @teubert,

Thanks a lot for your quick and thorough answer. Then I understand the purpose of this model and the way it works. It's fascinating reading about real use-cases.

Again, thanks for the clarification. I don't have any other questions for now but I'll be sure to come back if I do!