Real-Time Image-Based Fault Diagnosis for CHBMI-Fed IPMSM Drives Using a Multi-Branch CNN

This paper presents a fault diagnosis approach for a five-level Cascaded H-Bridge Multilevel Inverter (CHBMI) supplying an Interior Permanent Magnet Synchronous Motor (IPMSM). The method is based on a two-dimensional Convolutional Neural Network (2D CNN) designed to process voltage signals converted into image representations. In particular, the inverter voltage waveforms are transformed into grayscale images through a time-series reshaping procedure. This allows for the model to capture spatial patterns associated with different fault conditions, including both open-circuit and short-circuit faults. A multi-branch CNN architecture is adopted to simultaneously process multiple voltage signals, improving the ability to distinguish between fault types and locations. The proposed framework is evaluated on a dataset including 17 operating conditions under different speed and load profiles. The results confirm that the proposed approach provides accurate and reliable fault detection and is suitable for real-time diagnostic applications in multilevel inverter-based drive systems.