5.6-kW PM-SyRM, train current map, FEM data with spatial harmonics

This example trains a GradNet current map for a four-pole 5.6-kW PM synchronous reluctance machine (Baldor ECS101M0H7EF4) from a FEM dataset with spatial harmonics.

from pathlib import Path

import motulator.drive.gradnet as gn
from motulator.drive import utils

from motulator_gradnet import (
    Softmax,
    get_training_data,
    print_current_map_errors_fem,
    train_gradnet,
)

Compute base values based on the nominal values.

nom = utils.NominalValues(U=460, I=8.8, f=60, P=5.6e3, tau=29.7)
base = utils.BaseValues.from_nominal(nom, n_p=2)

Set up the paths and parameters.

p = Path(__file__).resolve().parent if "__file__" in globals() else Path.cwd()

dataset_path = p / "datasets/baldor_fem.npz"
trained_path = p / "trained_models" / "baldor_fem_curr_map_harm_softmax_d48_sub10.npz"
subsample = 10
k = 6
activation = Softmax

Train the model. The FEM data is on a uniform grid, so the loss is dominated by the saturated region. The loss of the data points whose current magnitude is below 12 A is weighted by 10 to improve the accuracy at low currents.

if not trained_path.exists():
    train_gradnet(
        dataset_path=dataset_path,
        base=base,
        save_model_path=trained_path,
        k=k,
        embed_dim=48,
        epochs=3000,
        subsample=subsample,
        activation=activation,
        i_low=12,
        weight_low=10,
    )

Load the dataset for visualization comparison.

# Get the training and validation data (complement) from the helper function
_, val_data = get_training_data(
    str(dataset_path), base=base, subsample=subsample, other_keys=["theta_m", "tau_m"]
)

Load the GradNet model and create its callable.

model = gn.load_gradnet(trained_path)
harm_map = gn.CurrentMapWithHarmonics(model, k=k)

Print statistical error metrics on validation data.

val_dict = dict(zip(["psi_s_dq", "i_s_dq", "theta_m", "tau_m"], val_data, strict=True))
print_current_map_errors_fem(map_fcn=harm_map, raw_data=val_dict, base=base)