Running external c-code function in MTK model

Specific Domains Modelling & Simulations
1

Hi community,

I am new to Julia and in particular MTK.
My goal is to simulate a battery cell and couple it with external C-code to run an extended Kalman filter for SOC estimation at a fixed sampling rate.

Somehow this kind of problem does not seem to be something many people deal with in Julia, at least documentation is very limited here. I guess FMU would be a possibility. However, the FMI.jl package hasn’t beed updated for a while and so far it does not support array inputs/outputs (which I ultimately need). Even then, how would the finite sampling time be realized, using events?

Is there another possibility to run external C-code functions using ccall and add the call at a certain rate to a symbolic system description? Ideally connecting it using inputs and outputs of the MTK system?

I would be grateful for general advice on best practices here.

Thanks in advance!

Edit: Just saw that FMI.jl has been updated 2 weeks ago.

2

Just register the function. See @register_symbolic

3

this is a discrete-time system and it requires the in-development support for synchronous programming

which is not quite available yet. You can still perform state estimation for MTK models in other ways, for example

4

Thanks, this seems to work!

5

That’s an interesting talk! Is there a timeline when this feature is to be published? How does this relate to tutorials → discrete_system (sorry I cannot add a link here)? Until the feature in the presentation is released, can I use events instead, specifically make ccalls in an event callback?

6

This is only for systems that are only discrete time, no continuous time dynamics.

Yes, it’s slightly cumbersome but not too bad if you have a single discrete-time component only

1 Like
7

Alright, i have a follow-up question.

I had a minimal example running where a continuous event would directly call a c function that was registered symbolic.

Now, my C-code wrapper has structures as inputs and outputs, i.e. I need to set those first, before doing the actual C-call.

This is my event inside @mtkmodel environment

@continuous_events begin
        [timer_wrapper ~ 1.0] =>
            [
                timer_wrapper ~ 0.0,
                soc_bat_est ~ wrapper_execute!(i_set, Tuple([Pre(getproperty(base_ecm, Symbol("cells_$i")).u_cg) for i in 1:n_series]))
            ]
    end

This is the Wrapper function

function wrapper_execute!(i_set::Real, u_cell::NTuple{NUM_CELLS, Real})
    inputs = WrapperInputs(current = Cfloat(i_set),
                           cellVoltages = ntuple(j -> Cfloat(u_cell[j]), NUM_CELLS))
    outputs = WrapperOutputs()
    retval = @ccall LIB_PATH.wrapper_execute(Ref(inputs)::Ptr{WrapperInputs},
                                           Ref(outputs)::Ptr{WrapperOutputs})::WrapperError

    return outputs.socMean
end

@register_symbolic wrapper_execute!(i_set, u_cell)

(instead of

wrapper_execute(val1, val2) = ... # ccall with val1 and val2 as doubles
@register_symbolic wrapper_execute(val1, val2)

)

Somehow, using the struct arguments this does not work, because the ccall in this case does not receive numeric values, but symbolic. What am I missing? Is there a workaround?

Cheers

8

Using a continuous event is probably suboptimal if you want a periodic execution, this is what discrete periodic callbacks are for. Having said that, you should hopefully be able to use what you have if you do

@register_symbolic wrapper_execute!(i_set::Real, u_cell::NTuple{NUM_CELLS, Real})