Estimation of Remaining Useful Life of Ball-bearings based on Complex Autoregressive Model of Double-axis Motor Current

This study aims to develop an easy-impement method for bearing health condition monitoring and predicting. Unlike traditional diagnostic methods with vibration sensor signals, this study utilizes motor current signature analysis (CSA) as a condition monitoring method to avoid cost and installation issues. Furthermore, the proposed anomaly detection approach can diagnose with just healthy samples, addressing the issue of insufficient fault samples in fault model training and classifying. This method introduced the complex autoregressive (CAR) model to identify the inherent pattern of health current. It can separate the nonsteady part in motor current caused by irregular vibration of damaged bearing. In addition, to enhance the robustness and accuracy of prognosis, this study uses the projected recursive least squares (P-RLS) for real-time prediction. It is shown to have better performance compared with traditional Bayesian-based methods by limiting the boundary of estimation. To verify the effectiveness of the proposed method, two different data sets are used for evaluation.