Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Data-Driven Model Identification of Unbalanced Induction Motor Dynamics and Forces using SINDYc

Emma Vancayseele, Philip Desenfans, Zifeng Gong, Dries Vanoost, Herbert De Gersem, Davy Pissoort

TL;DR

This work applies SINDYc to data-driven identification of unbalanced induction motor dynamics, focusing on stator currents in the dq0 frame as well as electromagnetic torque and unbalanced magnetic pull (UMP). Data are generated from a nonlinear magnetic equivalent circuit across three rotor eccentricity configurations, enabling accurate stator-current dynamics and improved torque/UMP mappings, though dynamic eccentricity poses greater challenges. The method avoids explicit flux estimation by using integrals of currents and voltages, and employs a randomized, Pareto-based hyperparameter search over multiple function libraries to yield sparse, interpretable models suitable for control. The results indicate enhanced torque modeling compared with conventional equations and high-accuracy UMP estimation for static eccentricity, with potential for predictive control of unbalanced drives and a path toward experimental validation in future work.

Abstract

This paper identifies the stator currents, torque and unbalanced magnetic pull (UMP) of an unbalanced induction motor by the System Identification of Nonlinear Dynamics with Control (SINDYc) method from time-series data of measurable quantities. The SINDYc model has been trained on data coming from a nonlinear magnetic equivalent circuit model for three rotor eccentricity configurations. When evaluating the SINDYc model for static eccentricity, torques and UMPs with excellent accuracies, i.e., 8.8 mNm and 4.87 N of mean absolute error, respectively, are found. When compared with a reference torque equation, this amounts to a 65% error reduction. For dynamic eccentricity, the estimation is more difficult. The SINDYc model is fast enough to be embedded in a control procedure.

Data-Driven Model Identification of Unbalanced Induction Motor Dynamics and Forces using SINDYc

TL;DR

This work applies SINDYc to data-driven identification of unbalanced induction motor dynamics, focusing on stator currents in the dq0 frame as well as electromagnetic torque and unbalanced magnetic pull (UMP). Data are generated from a nonlinear magnetic equivalent circuit across three rotor eccentricity configurations, enabling accurate stator-current dynamics and improved torque/UMP mappings, though dynamic eccentricity poses greater challenges. The method avoids explicit flux estimation by using integrals of currents and voltages, and employs a randomized, Pareto-based hyperparameter search over multiple function libraries to yield sparse, interpretable models suitable for control. The results indicate enhanced torque modeling compared with conventional equations and high-accuracy UMP estimation for static eccentricity, with potential for predictive control of unbalanced drives and a path toward experimental validation in future work.

Abstract

This paper identifies the stator currents, torque and unbalanced magnetic pull (UMP) of an unbalanced induction motor by the System Identification of Nonlinear Dynamics with Control (SINDYc) method from time-series data of measurable quantities. The SINDYc model has been trained on data coming from a nonlinear magnetic equivalent circuit model for three rotor eccentricity configurations. When evaluating the SINDYc model for static eccentricity, torques and UMPs with excellent accuracies, i.e., 8.8 mNm and 4.87 N of mean absolute error, respectively, are found. When compared with a reference torque equation, this amounts to a 65% error reduction. For dynamic eccentricity, the estimation is more difficult. The SINDYc model is fast enough to be embedded in a control procedure.

Paper Structure

This paper contains 11 sections, 19 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Theory
  3. Identification of Nonlinear Dynamical Systems
  4. Identification of Nonlinear Mappings
  5. Methodology
  6. Selection of State and Input Variables
  7. Data-Generation Model
  8. Results and Discussion
  9. Hyperparameter Optimisation
  10. Evaluation of Performance and Sparsity
  11. Conclusion

Figures (4)

  • Figure 1: Validation of data-generation model for steady-state motor operation.
  • Figure 2: Visualisation of an optimisation study for the stator current dynamical model.
  • Figure 3: Evaluation of SINDYc model performance for stator current, unbalanced magnetic pull, and electromagnetic torque modelling. The data-generation model is visualised using the black dotted line as a reference.
  • Figure 4: Visualisation of SINDYc weighting matrices. The elements in white are precisely zero.