Predication of remaining useful life of aircraft engines based on Multi-head Attention and LSTM

Abstract

According to the large amount of high-dimensional time series data generated by sensors, and the insufficient utilization of time series information in the network model in the prediction on remaining useful life (RUL) of aircraft engines, a data-driven model for RUL prediction based on a Multi-head Attention mechanism and a long short-term memory neural network (LSTM) is proposed in this paper to optimize RUL of aircraft engine. The model can select the key features in the time-series data, then input them into the LSTM layer to mine the internal connections, and finally obtain RUL predicted results through two fully connected layers. Using the CMAPSS dataset provided by NASA for verification and comparing with other algorithms, the accuracy of this method outperforms shallow neural networks based on support vector regression (SVM) and deep learning methods such as convolutional neural networks (CNN), multi-layered LSTM and multi-layered BiLSTM, which provides a powerful support for health management of aircraft engines, operation and maintenance decisions.