01 / Context
Overview
This work studies automatic detection and localization of open-circuit faults in multiphase induction drives. The central idea is to learn the relationship between drive operating variables and fault classes using neural networks, while retaining enough variation in speed and loading for the classifier to generalize beyond a single operating point.
Simulation was used to generate large labeled datasets efficiently, while experimental measurements from the five-phase drive platform were used to test whether the learned classifier transferred to real hardware. The research therefore combines physics-based drive models, data generation, machine learning, and experimental validation.
My work
- Generated labeled normal, open-phase, and open-switch datasets across drive operating conditions.
- Trained neural-network classifiers for fault detection and localization.
- Fine-tuned/tested the models on experimental data from the laboratory drive system.
- Used the results to support future fault-tolerant-control development.
02 / Approach
Methods & diagrams
Data Generation
Drive currents, speed, and related signals were collected under healthy and faulted conditions using simulation and hardware experiments.
Fault Classification
Feedforward neural networks learned fault signatures without requiring manually engineered features for every fault class.
Experimental Validation
Hardware datasets were used to assess whether models trained from simulation retained useful performance on the physical system.
03 / Evidence
Results & gallery
Project figures and laboratory photographs. Open a figure to inspect the detail; vector PDFs are available for the control diagrams.
Findings
- The project demonstrated a practical simulation-to-experiment workflow for data-driven drive diagnostics.
- Multiphase drives provide distinctive phase-current fault signatures that can be exploited for localization as well as detection.
- Experimental testing was essential for evaluating generalization beyond idealized simulation data.
Scope & limitations
The figures illustrate the documented normal and faulted operating cases. Diagnostic performance is specific to the dataset, fault classes, and evaluation protocol.
