Transfer between multiple working conditions: A new TCCHC-based exponential semi-deterministic extended Kalman filter for bearing remaining useful life prediction

Traditional remaining useful life (RUL) prediction methods developed in ideal environment are not applicable in real industrial world. This paper presents a new approach that combines transfer compact coding for hyper plane classifiers (TCCHC) with exponential semi-deterministic extended Kalman filter (EKF) to transfer the RUL prediction models among multiple working conditions, where three major processes are involved: Mel-frequency cepstral coefficient (MFCC) process for degradation curve establishment, TCCHC process for transfer learning, and exponential semi-deterministic EKF process for bearing RUL prediction. Here the main principle of transfer learning is to select and transfer the MFCC degradation curve from one working condition to another working condition. Furthermore, the purpose of exponential semi-deterministic EKF model is to obtain the probability density distribution of the RUL. Related experiments proved that transfer strategy has a significant advantage especially under varying working conditions, thus being a useful tool for bearing RUL prediction in real industrial system.