Volume 08,Issue 04

A Hybrid CNN-BiLSTM and Wiener Process-based Prediction Approach of Remaining Useful life for Rolling Bearings

Authors

Jia-Lun Wan, Yan Yang, Junyu Guo , Le Dai

Abstract
Predicting the remaining useful life (RUL) of rolling bearings can provide guidance and reference for effective maintenance of rolling bearings in advance to ensure the regular operation of the machine. Therefore, maintaining bearings’ secure and reliable work is of great significance. Toward this end, this paper presents a RUL prediction model based on a convolutional neural network (CNN) and bidirectional long short-term memory (BiLSTM) hybrid neural network, combined with the Wiener process. This method has two components: extraction of vibration signal features based on the CNN-BiLSTM model and RUL prediction of bearings using the Wiener process. Since the technique of constructing feature engineering after dimensionality reduction of the time-frequency features of the bearing may lose important signal information, thus, this paper tries to use the vibration acceleration signal of the bearing as the input feature and then use CNN and BiLSTM to build the bearing degradation model. Health index construction by the advantages of CNN for feature extraction and BiLSTM for processing sequence data. Considering the uncertainty of the bearing degradation process, finally, the Wiener process is used to deduce the probability density function (PDF) for predicting RUL to predict the RUL of the constructed health index model. The PHM 2012 bearing datasets confirm the validity and superiority of the presented method in this study.

Keyword: RUL, CNN, BiLSTM, Wiener process, Bayesian method.

PDF [ 809.74 Kb ] | Endnote File | XML