Remaining useful life prediction of the aircraft engine based on the GRU-GAN network with a feature attention mechanism

Abstract

The turbofan engine is an important piece of equipment in the aerospace industry, and its health directly affects the aircraft's stability and reliability. The remaining useful life of the turbofan engine is an important factor for aircraft monitoring and maintenance. However, the existing monitoring method has a number of shortcomings, such as complex operating conditions, redundant and largescale volume data, and a long time span, all of which contribute to a less accurate estimate of the remaining useful life. This paper proposes a fusion of Generative Adversarial Networks (GAN) and Gated Recurrent Units (GRU) based on the mechanism of feature attention, combining the generation capability of GAN with the predictive capability of GRU. To develop the time-series prediction model, feature attention mechanism and GRU modules are used to extract spatial and temporal correlations. The final prediction is made using a pre-trained GAN generator connected to the GRU's output. Specifically, substituting the typical encoder-decoder network with a GAN resolves the complicated and redundant data problem, significantly boosting the model's prediction precision and generalization performance. This article verifies the model effect using CMAPSS turbofan engine data and compares it to machine learning methods. The experimental results indicate that the method has good predictive accuracy for two evaluation indicators: the root mean square error and the score.