Inverter Open Circuit Fault Diagnosis Method Based on Residual Evaluation and Machine Learning

To address the issues of limited noise resistance capability in fault diagnosis and inherent difficulties in fault isolation within conventional model‐driven methods, an inverter open‐circuit fault diagnosis method based on residual evaluation and machine learning is proposed in this paper. Firstly, the linear Kalman filter is designed to estimate three‐phase currents subject to the noise interference. Current residuals are calculated by subtracting the estimated values from actual fault currents. Secondly, the residual evaluation function based on the p‐norm is introduced. Subsequently, the residual evaluation features within each time subdomain are computed in sequence. Meanwhile, the alarm threshold via weight negative correlation is put forward. Fault diagnosis is achieved by comparing evaluation features with the threshold. Thirdly, the feedforward neural network is employed to extract nonlinear features of current residuals. Feature components after principal component analysis are input into the support vector machine to accomplish classification tasks and locate failure power transistors. Simulation results show that the fault diagnosis scheme performs the fast diagnosis speed and has the ability to avoid false alarms. The fault isolation scheme based on the machine learning framework achieves precise location of the failure power transistor on the basis of rapid diagnosis.