McMaster University
CURRENT-SIGNATURE-BASED DEEP LEARNING FRAMEWORKS FOR BEARING AND OPEN-CIRCUIT FAULT DIAGNOSIS IN PERMANENT MAGNET SYNCHRONOUS MOTORS (PMSMS)
Abstract
dc:description.abstractPermanent magnet synchronous motors (PMSMs) are widely used in electrified systems, including electric vehicles and industrial applications, due to their high efficiency and power density. However, faults such as bearing defects and open-circuit faults (OCFs) can degrade performance, leading to increased vibration, torque ripple, and reduced reliability. Early and accurate fault diagnosis is therefore essential for safe and cost-effective operation. This thesis presents current-signature-based diagnostic frameworks and mathematical modeling for PMSM bearing faults and OCFs. The diagnostic methodology is first developed and validated for bearing faults and subsequently extended using tailored techniques for OCF diagnosis, with real-time implementation demonstrated. For bearing faults, a multilevel cascaded artificial neural network (ANN) framework is developed for fault severity classification. Multi-resolution features are extracted from stator current signals using the discrete wavelet transform (DWT) and fused with measured operating parameters to improve diagnostic accuracy and reliability under varying operating conditions. A risk-based performance metric, along with an optimal model selection criterion, is proposed to evaluate diagnostic performance while balancing accuracy and computational complexity. For OCFs, a dedicated diagnostic framework is developed using fused multi-domain features derived from wavelet analysis, high-frequency (HF) injection, frequency-domain characteristics, and speed information. These features are processed using a lightweight deep neural network (DNN) with an additional refinement layer to improve fault detection and localization. A transfer learning-based approach is employed to facilitate domain adaptation from the simulation domain to the experimental domain, enabling reliable hardware-based validation. The proposed risk-based performance metric is further extended to enable consistent evaluation across fault types. A comprehensive validation framework integrates PMSM dynamometer experiments with real-time hardware-in-the-loop (HIL) implementation to assess performance under varying speed and load conditions. Sensitivity analysis evaluates the robustness of features under HF injection and varying operation. A real-time HIL platform based on a C2000 microcontroller validates embedded feasibility through execution-time profiling, with detection and localization latency quantified in electrical fundamental frequency cycles. The proposed frameworks eliminate the need for additional sensors and enable integration into existing PMSM drive systems. Extensive validation demonstrates diagnostic accuracies exceeding 99% across a wide range of operating conditions, with low detection and localization latency for open-circuit faults and strong resilience to noise and operational uncertainties, particularly in bearing fault diagnosis. Hardware validation confirms reliable real-time performance for OCF diagnosis under practical deployment constraints.
Degree
thesis:*- Department dc:contributor.department
- Electrical and Computer Engineering
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Jayasena,Korawege Nimesh Chathuranga
- Advisor dc:contributor.advisor
-
- Emadi, Ali
Subjects
dc:subject × 24- Permanent Magnet Synchronous Motors (PMSMs)
- Current Signature Analysis (CSA)
- Current-Signature-Based Fault Diagnosis
- Condition Monitoring
- Fault Diagnosis
- Bearing Fault Diagnosis
- Bearing Fault Severity Classification
- Open-Circuit Fault Diagnosis
- Deep Neural Networks (DNNs)
- Feature Extraction
- Feature Fusion
- Discrete Wavelet Transform (DWT)
- Transfer Learning
- Domain Adaptation
- Real-Time Fault Diagnosis
- Hardware-in-the-Loop (HIL)
- Co-Simulation
- High-Frequency Injection
- Physics-Informed Fault Modeling
- Risk Distribution Matrix
- Risk-Weighted Performance Indicator
- Classification Refinement Layer
- Multilevel Cascaded Deep Learning Framework
- Noise Robustness Evaluation
Rights
dc:rights- Statement dc:rights
-
- Attribution-NonCommercial-NoDerivatives 4.0 International
- Licence dc:rights.uri
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Identifier URI
- https://doi.org/10.71548/2210