Remaining useful life estimation model for aero-engine using multi-feature attention

Abstract

The degradation trend of aero-engine performance is complex, so it is very important to predict its remaining life and maintain it in time. In this paper, a dilated convolution network based on multi-feature attention model is presented to predict the remaining useful life (RUL) of aero-engine. In this model, dilated convolution is used to enhance the ability to extract temporal features of sequence data, and residual connections are established to improve the problem of gradient disappearance in traditional convolution networks. Firstly, the raw input data are reconstructed by sliding time window of fixed length to intercept data along the time dimension. Then the dilated convolution networks extract the temporal features of corresponding to each feature respectively. Finally, the feature attention mechanism is used to calculate the relative importance of features. Experimental results demonstrate that the proposed algorithm has better accuracy of RUL estimation than the other comparative models.

Keywords

• aero-engine
• dilated convolution
• residual connection
• multi-feature attention
• remaining useful life estimation