Research on Prognosis for Engines by LSTM Deep Learning Method

Abstract

Prognostics and health management (PHM) technology has been successfully applied in many complex equipment. However, with the equipment becoming more and more complex, the working conditions changing with time, and the equipment status information increasing, it is difficult by traditional technologies to cope with the new situation and new application scenarios. The application of deep learning method in many fields proves the ability of this method to deal with massive and complex data. In this paper, the special recurrent neural networks (RNN) called long-short term memory (LSTM) network are used to estimate the remaining life of engines with the data of PHM08 Challenge Competition. First, standardize the original data and add life labels in the data preprocessing stage. Then the influences of different data input methods on the prediction results are studied, and the results show that proper method is to input all the time series information at one time. The over-fitting phenomenon can be reduced to some extent by reducing the complexity of the neural network. Thus, a remaining life prediction method based on multi-dimensional data is obtained. The final result was uploaded to the competition’s scoring system and got good results, which confirmed the accuracy of this method. Therefore, the article summarizes a highly accurate LSTM-based multidimensional data failure prediction method.