Multiphysics modelling-based digital twin for intelligent fault diagnosis of PMSMs

Permanent magnet synchronous motors (PMSMs) are key actuators in modern industrial and transportation systems, where reliable fault diagnosis is essential to ensure safe and efficient operation. However, most existing PMSM models are confined to a single physical domain and lack support for customizable fault injection, limiting their accuracy and applicability in f ault diagnosis tasks. To address these limitations, this study presents a digital twin (DT)-driven intelligent diagnostic framework of PMSMs. First, a high-fidelity multiphysics DT model of the PMSM, coupling electromagnetic, mechanical, and thermal domains, is developed in Modelica. Subsequently, physically consistent, temporally resolved fault datasets are generated by simulating representative fault scenarios. Finally, a convolutional neural network (CNN)-based, structure-aware deep learning architecture is trained for accurate fault classification and onset detection via a temporal windowing mechanism. On the validation set, the model achieves 98.08% accuracy and near-real-time performance (onset latency ≈ 10–20 ms). The results demonstrate superior performance in fault data generation, diagnostic accuracy, and model interpretability, providing strong support for the development of reliable and deployable DT systems for electric machines. This approach enables the industrial-scale application to electric machines across the transportation and energy sectors, facilitating real-time monitoring and enhancing operational reliability.