Hybrid Fault Diagnosis of Multiple Open-Circuit Faults for Cascaded H-Bridge Multilevel Converter Based on Perturbation Estimation Convolution Network

This article proposes a hybrid fault diagnosis method based on perturbation estimation convolution network (PECN) of multiple open-circuit switch faults for cascaded H-bridge (CHB) multilevel converter. The proposed perturbation observer as the model-based method can extract fault characteristics of output current and voltage. The deviations of measured states and observed states, which are introduced as perturbation estimation, as well as capacitor voltages form the input data of convolution neural network (CNN). The multilayer convolution network is applied to deeply extract the fault signatures and determine the type and location of faulty switches rather than manually setting empirical thresholds as in the traditional model-based methods. The proposed PECN method improves accuracy and adaptability through combining the advantages of both model-based and data-driven method, which can detect and locate multiple open-circuit faults under different faults and operation conditions. Simulations results confirm the effectiveness and robustness of the proposed PECN method, which are further demonstrated on a hardware-in-the-loop (HIL) testing platform.