Remaining useful life prediction for machinery by establishing scaled-corrected health indicators

Huge amount of data generated from condition monitoring can be used for remaining useful life (RUL) prediction, which is not only the prerequisite for predictive maintenance but also a relevant function in smart manufacturing systems. However, finding a suitable health indicator (HI) representing the degradation trend from multi features is still the difficult point in this issue. In addition, the variety of failure threshold under different operating conditions hinders the application of the existing methods. This paper proposes an efficient method for the RUL prediction with the scaled-corrected HI, which is constructed through unsupervised learning. A scaling parameter is introduced to unify the different failure threshold. The sensitive features are selected from multiple domain parameters and the RULs are estimated by the particle filter. The effectiveness of the proposed method is experimentally validated through a bearing dataset and compared with the state-of-the-art.