An Intermittent Fault Diagnosis Method for Multiple Sensors Based on Multi-AI Predictors in Induction Motor Drive System

In this article, the problem of multiple sensor's intermittent faults (IFs) diagnosis is investigated for induction motor drive systems. To overcome the difficulties in locating IFs for different sensors, a set of AI predictors trained by a pair of homologous and heterogeneous datasets is built to construct residuals only sensitive to their corresponding sensor. As components of a predictor, nonlinear autoregressive exogenous (NARX) structure, random vector functional link (RVFL), and linear feedback structure are used to extract the mapping relationships between inputs and outputs. Then, based on the sliding window and the unique characteristics of predictors, a pair of decoupled residuals is constructed. After that, a threshold-based decision-making method is designed to identify the dc-link voltage sensor IFs and phase current sensor IFs. Finally, the robustness and effectiveness of the proposed IFs diagnosis method are validated in a 3.3 kW experimental platform with a three-phase induction motor drive system.