Remaining Useful Life Prediction of Rolling Element Bearings Based on Hybrid Drive of Data and Model

Abstract

For remaining useful life (RUL) prediction of machinery, model-driven methods often use a single model to process individual data, which is difficult to adapt to the diversity of degradation behaviors. Data-driven methods are more dependent on training data, and in practice a large amount of run-to-failure data is difficult to obtain. In this paper, a new hybrid drive of data and model method is proposed. In the model-driven path, a new scalable two-stage linear/nonlinear composite model is constructed to represent various degradation behaviors, and to clarify the evolution law of individual degradation. In the data-driven path, the long short-term memory prediction network is trained to track the degradation process and learn knowledge of multi-sample degradation behavior. The newly established dynamic matching index integrates the model-driven and data-driven paths, and realizes the interactive fusion of information and RUL prediction through real-time matching of hidden layer states. The whole life cycle performance degradation data of two sets of different experimental rigs are used for analysis, and compared with some state-of-art RUL prediction methods, the results show that the proposed method has higher prediction accuracy.