Abstract:

To solve the problem of low accuracy of remaining useful life prediction due to insufficient monitoring data of the whole life cycle of rolling bearings, a method of remaining useful life prediction based on time series data augmentation and Bidirectional long-term short term memory network (BLSTM) is proposed. Firsty, the data of the vibration acceleration of the training rolling bearing and the test bearing are collected. Secondly, after preprocessing the collected raw data, the health factors are extracted, and the training data and test data are formed into reference data set and target data set respectively. Thirdly, based on the reference data set, the target data set is augmented by the dynamic time warping algorithm.Finally, the BLSTM network is trained with the augmented test data, and the trained BLSTM network is used to predict the performance degradation trend and remaining useful of the rolling bearing.The experimental results show that the data augmentation model based on dynamic time warping algorithm can effectively improve the prediction ability of performance degradation trend and the prediction accuracy of remaining useful life by training BLSTM network based on the existing rolling bearing operation data with similar amplification performance degradation process based on the existing full-life cycle data.

Key words: time series data, data augmentation, dynamic time warping, remaining useful life prediction, long-short term memory (LSTM) network