A dual path neural network based on spatial correlation and feature weighting for predicting remaining useful life of aero-engine
点击次数：
DOI码：10.1016/j.engappai.2025.110700
发表刊物：Engineering Applications of Artificial Intelligence（中科院一区）
摘要：Accurate prediction of the remaining useful life (RUL) of aero-engine is crucial for flight safety, the reduction of maintenance costs and the extension of engine service life. Despite that deep learning methods have achieved favorable outcomes in the field of RUL prediction in recent years, the spatial correlations among feature information have not been adequately extracted and insufficient attention has been given to the variations in the importance of different input features across various time steps. To tackle these problems, a novel hybrid dual path network prediction method based on Multi-scale Convolution Neural Network (MS-CNN) and Gated Recurrent Unit (GRU) is proposed in this article. On the one hand, an innovative high-dimensional feature correlation extraction method is developed, multiple graph structures are constructed to capture complex spatial relationships among different feature maps formed by MS-CNN. On the other hand, under the consideration of the impact of the leaning of previous data in GRU on subsequent input features, feature attention mechanism is integrated with the hidden state information in GRU network to achieve differential update of input features at different time steps. The effectiveness of proposed model is validated on two well-known datasets and compared to other methods with similar neural network structures. And the experimental results show that the proposed framework could achieve more accurate RUL prediction than existing methods, especially under complex conditions and multiple fault modes.
合写作者：Wanlin Liu（通讯作者）,Ba-Yi Cheng
第一作者：Mi Zhou
论文类型：期刊论文
学科门类：管理学
文献类型：J
卷号：151
页面范围：110700
是否译文：否
发表时间：2025-04-01
收录刊物：SCI