InductionMachineDrives.jl

Julia package for modelling, simulating, and testing induction-machine control strategies, providing reusable machine models, observers, discrete controllers, estimators, and standalone benchmark simulators for a squirrel-cage induction machine under field-oriented control (FOC).

The package contains both continuous ModelingToolkit systems and discrete/hybrid controller implementations. It is also used as the machine-control backend for ElectricMachineWinch.jl, which integrates the induction-machine model inside a KiteControllers.jl autopilot simulation.

As example, simulation scripts for the AWES ground-station electrical bench of "Universidad Carlos III de Madrid" are provided.

Two modelling paths

The package deliberately keeps two independent paths side by side.

Hybrid FOC — the package core, and where the current work happens. A hand-written induction-machine plant is stepped alongside hand-written discrete controllers and observers on a fixed-step loop, which is close to how a real embedded implementation behaves. It is plain Julia over concrete Float64, with no symbolic types anywhere.

Scalar V/f and symbolic FOC — ModelingToolkit systems used mainly for plant validation. Everything symbolic lives in a package extension, not in src/.

Each fixed-step iteration of the hybrid loop does:

read plant currents / speed / angle
update rotor-flux and torque observer
optionally update load-torque Kalman estimator
update outer loop (torque or speed), if present
update inner current controller
convert the dq voltage command to alpha-beta
integrate the plant one sample period
store results

Mechanical sign convention

This one convention runs through the plant, the estimators, and every feedforward path:

\[J \frac{d\omega}{dt} = T_{\mathrm{e}} + T_{\mathrm{L}} - B \omega\]

$\omega$ — mechanical speed of the machine, rad/s
$T_{\mathrm{e}}$ — electromagnetic torque, Nm
$T_{\mathrm{L}}$ — external or load torque, Nm
$B \omega$ — viscous friction, opposing motion

Both $T_{\mathrm{e}}$ and $T_{\mathrm{L}}$ are positive when they accelerate the shaft toward positive speed; a positive $T_{\mathrm{L}}$ pulls the shaft in the positive direction rather than resisting it.

Getting this backwards fails silently

A flipped sign here does not throw — it quietly breaks load feedforward and the Kalman estimator. Before changing feedforward or estimator code, cross-check sections 3, 9 and 11 of the architecture page, which spell out the required signs (load_ff_sign = -1.0, Kalman A12 = Ts/J rather than -Ts/J, and TL_kalman_limit_positive = false for AWES profiles carrying signed torque).

Loading the package

Julia 1.12 is required. The two MTK-backed model builders are gated behind a package extension:

using InductionMachineDrives                                    # hybrid FOC simulators only
using ModelingToolkit, OrdinaryDiffEq, InductionMachineDrives   # + the symbolic system builders

Both triggers are needed. Julia loads an extension only once every package in its trigger list is present, and OrdinaryDiffEq is not optional — the simulate_* functions default to Rodas5P() and call solve.

A `MethodError` with no methods listed is the expected signal

build_scalar_im_system, simulate_scalar_im, build_foc_current_im_system and simulate_foc_current_im are empty stubs until the extension loads. Calling one after using ModelingToolkit alone raises a MethodError listing zero methods. That means a trigger is missing, not that the function is broken.

The package's own [deps] is empty by design, so using InductionMachineDrives on its own pulls in nothing.

Getting started

Clone the repository and set up the workspace:

bin/install       # checks Julia 1.12, restores the known-good manifest,
                  # instantiates the workspace, precompiles scripts/
bin/run_julia     # REPL with the right project and environment variables

In that REPL, menu() lists the hybrid FOC examples and menu2() the ModelingToolkit ones. The split is deliberate: only the menu2() scripts pull in ModelingToolkit and OrdinaryDiffEq.

A single script can also be run directly — each activates the scripts/ environment itself:

julia scripts/run_foc_current_hybrid_steps.jl

Scripts write CSV output to results/ and plot with MakieControlPlots.

Set `KMP_DUPLICATE_LIB_OK`

bin/install and bin/run_julia export KMP_DUPLICATE_LIB_OK=TRUE. If you start Julia any other way, set it yourself — especially on Windows.

AWES reference profiles

Input profiles such as profiles/delta_kite_13_ms_profiles.csv carry the columns t_s, speed_ref_rpm, speed_ref_rad_s and torque_ref_Nm. They are linearly interpolated at each simulation timestep, matching how a Simulink timeseries source behaves.