A Hybrid Model and Data-Driven Approach for Online Diagnosis of Open-Circuit Faults in Grid-Tied Three-Phase VSIs

This paper proposes a hybrid method for diagnosing single and multiple transistor open-circuit faults in grid-tied three-phase voltage source inverters. Combining explicit variable relationships in analytical models with the nonlinear regression capability of neural networks, the method comprises offline model training and online fault diagnosis sections. The offline section constructs a neural network model based on analytical model variables, using closed-loop system samples to predict fault characteristics. In the online part, the predictive model is applied to the Simulink online simulation platform. Real-time predictions and auxiliary signals enable online diagnosis of open-circuit faults, ensuring rapid diagnosis without additional hardware circuitry, addressing challenges like computational intensity, difficult threshold selection, and complex rule formulation. Simulation results validate the method's excellent performance.