Open-Circuit Fault Diagnosis for Three-Level ANPC Inverter Drives Using Improved 1D Convolutional Neural Network

Multilevel inverters (MLIs) have been drawn attractive attention in medium-voltage high-power industry applications due to their lower dv/dt and harmonics. The three-level active neutral-point-clamped (3L-ANPC) inverter displays small switching losses and uniform loss distribution. However, the increased number of switching transistors implies a higher probability of switch open-circuit (OC) faults. To ensure the stable and secure operation of the system, an OC fault diagnosis method based on an improved 1D convolutional neural network (CNN) is proposed, where the fully connected layer (FC) is replaced with a global max pooling layer (GMP), and the multi-channel input is employed. The collected current data are put into the CNN model for training, and it can diagnose the typical fault cases with the average accuracy of 99.7%. Finally, experiments are conducted on a platform of 3L-ANPC inverter fed permanent magnet synchronous motor (PMSM) drives to verify the effectiveness and robustness of the proposed method.