Enhanced 1-D Convolutional Neural Network-Based Open-Circuit Fault Diagnosis and Hybrid Fault-Tolerant Control for Three-Level NPC Converters

This article addresses the challenge of diagnosing open-circuit faults in power devices within three-level neutral point clamped (NPC) converters. An enhanced 1-D convolutional neural network (1D-CNN)-based fault diagnosis method is proposed. The method begins with the acquisition of fault data, including three-phase voltages on the dc side and three-phase output currents. A fault feature matrix is constructed using the complementary ensemble empirical mode decomposition (CEEMD) algorithm, followed by energy percentage feature extraction. This matrix is then processed by the enhanced 1D-CNN framework, which effectively detects both single and dual open-circuit faults in power devices. To address single open-circuit faults, a fault-tolerant control strategy based on space vector pulsewidth modulation (SVPWM) is introduced, ensuring continuous operation by adjusting the space vector upon fault detection. In addition, a redundant power unit is employed to maintain consistent output voltage and current amplitude during fault-tolerant operations. The proposed method’s effectiveness is validated through simulation and experimental results, demonstrating its capability to accurately locate faulty devices and enable fault-tolerant operation. This research presents a reliable solution for open-circuit fault diagnosis, improving the resilience and operational efficiency of three-level NPC converters.