Research team:

• Prof. Teresa Orłowska-Kowalska, PhD, DSc, Eng (project manager),
• Marcin Wolkiewicz, PhD, DSc, Eng,
• Grzegorz Tarchała, PhD, DSc,  Eng,
• Paweł Ewert, PhD, Eng,
• Tomasz Zawilak, PhD, MSc, Eng,
• Maciej Skowron, MSc, Eng,
• Mateusz Krzystofiak, MSc, Eng,
• Przemysław Pietrzak, MSc, Eng.

The aim of the project was to develop new methods for the detection and diagnosis of electrical and mechanical faults in drive systems with permanent magnet synchronous motors using advanced signal processing methods and neural networks. As part of the project, hybrid and direct systems for detecting and classifying basic faults in PMSM motors operating in a closed-loop vector control structure, using machine learning methods, were developed and tested through simulation and experimental studies. Diagnostic systems were developed based on deep (convolutional) networks that directly process signals obtained from measurements or simulation studies of the drive, as well as hybrid systems using various signal processing methods and shallow neural networks (e.g., multilayer perceptron, Kohonen network, cascade networks) or various classification algorithms (e.g., K-nearest neighbors, support vector method, decision tree, naive Bayes classifier).

The fundamental research conducted within the project focused on five main topics:

1. Development of mathematical models for simulating electrical winding and permanent magnet faults in PMSM motors, and generation of turn short circuit and demagnetization symptoms based on simulation studies.
2. Preparation of a test bench with a PMSM motor enabling physical modeling of electrical (short turns), magnetic (permanent magnet damage), and mechanical (bearing damage, misalignment, rotor imbalance) faults, generating fault symptoms, and verifying developed signal processing methods and neural fault detectors and classifiers.
3. Analysis, development, and testing of classic and advanced algorithms for processing current and vibration acceleration signals through simulation and experimental studies, enabling the generation of electrical and mechanical fault symptoms for AC motors, particularly PMSM motors.
4. Development and experimental testing of hybrid PMSM motor fault detectors and classifiers based on fault symptoms obtained from pre-processed measurement signals and on shallow neural networks and classification algorithms.
5. Development and experimental testing of PMSM motor fault detectors and classifiers based on convolutional networks trained on raw measurement data or simulation results and verified in a real drive.

The research resulted in intelligent diagnostic systems capable of detecting electrical, magnetic, and mechanical faults in their initial stages, as well as classifying the level or type of selected faults. The proposed solutions have implementation potential and therefore can be an alternative to previously used diagnostic techniques in both industrial and commercial applications. The results of the work were presented at international conferences (PEMC, EDPE, ELECTRIMACS, SPEEDAM) and national conferences (SENE). The most valuable results of the work have been published in international journals (IEEE Transactions on Industrial Electronics, IET Electric Power Applications, IEEE Access, Bulletin of the Polish Academy of Sciences – Technical Sciences, Archives of Electrical Engineering, Electronics, Energies, Power Electronics and Drives).

Some of the research results carried out within the project were included in the defended doctoral theses of the project scholarship holders (Maciej Skowron, "Diagnosis of faults in induction and synchronous motors with permanent magnets using neural networks with deep learning" – 2021, Mateusz Krzysztofiak, "Mathematical modeling methods in fault diagnosis of synchronous motors with permanent magnets" – 2024 and Przemysław Pietrzak, "Diagnosis of faults in stator windings in drive systems with synchronous motors with permanent magnets using advanced signal processing and artificial intelligence methods” – 2025.