A method for predicting the remaining life of equipment based on WTTE-CNN-LSTM

Abstract

At present, the uncertainty and randomness between equipment are not fully considered in the remaining useful life (RUL) prediction. In order to solve this problem, firstly, we use the Weibull distribution to describe the influence of various uncertain factors on the RUL of equipment, and introduce the Weibull Time-To-Event Recurrent Neural Network (WTTE-RNN) framework to transform the RUL of equipment from the prediction of single life value to the prediction of Weibull distribution parameters. Then, in view of the problem that RNN is prone to have low prediction accuracy due to the vanishing of gradient, considering the advantages of Long-Short Term memory (LSTM) in time series modeling, we replace RNN with LSTM to improve the model and construct WTTE- LSTM model. Furthermore, in order to further improve the model's ability to extract data features, Convolutional Neural Network (CNN) is added after the original data is normalized because of its excellent feature extraction ability, and the time series features extracted by CNN are used as the input of LSTM to construct the WTTE-CNN-LSTM model. Finally, the LSTM life prediction model, WTTE-LSTM model and WTTE-CNN-LSTM model are established by taking a data set from a core component of construction machinery as an example. The results demonstrate that the improved WTTE-CNN-LSTM model has the highest prediction accuracy and the smallest error.

Keywords

• remaining useful life prediction(rul)
• weibull time-to-event recurrent neural network (wtte-rnn)
• convolutional neural network (cnn)
• long-short term memory (lstm)