基于融合卷积Transformer的航空发动机故障诊断  

Aero-engine fault diagnosis based on fusion convolutional Transformer

作　　者：赵洪利[1] 杨佳强 ZHAO Hongli;YANG Jiaqiang(School of Aeronautical Engineering,Civil Aviation University of China,Tianjin 300300,China)

机构地区：[1]中国民航大学航空工程学院,天津300300

出　　处：《北京航空航天大学学报》2025年第4期1117-1126,共10页Journal of Beijing University of Aeronautics and Astronautics

基　　金：中央高校基本科研业务费专项资金(3122021049);中国民航大学实验技术创新基金(2021CXJJ90);2022年天津市研究生科研创新项目(2022SKY156);中国交通教育研究会2022—2024年度教育科学研究课题(JT2022YB326)。

摘　　要：航空发动机长期处于恶劣的气路环境下工作会面临腐蚀、侵蚀等问题,且故障参数特征不明显,因此,精准的航空发动机故障诊断方法对保证飞机安全运行具有重要意义。为提高预测准确性,提出了一种基于融合卷积Transformer的航空发动机故障诊断方法。利用自注意力机制提取有用特征,抑制冗余信息,并将最大池化层引入Transformer模型中,进一步降低模型内存消耗及参数量,缓解过拟合现象。采用基于GasTurb建模的涡扇发动机仿真数据集进行验证,结果与Transformer模型和反向传播(BP)神经网络、卷积神经网络(CNN)、循环神经网络(RNN)等传统深度学习模型相比,准确率分别提高了6.552%和28.117%、13.189%、10.29%,证明了所提方法的有效性,可为航空发动机故障诊断提供一定的参考。Aero-engine faces the problems of corrosion and erosion when working in an atmospheric environment for a long time,and the features of fault parameters are not obvious.Therefore,an accurate aero-engine fault diagnosis method is of great significance to ensure the safe operation of aircraft.To improve the prediction accuracy,an aero-engine fault diagnosis method based on a fusion convolutional Transformer was proposed,which used the self-attention mechanism to extract useful features and restrain redundant information.In addition,the MaxPool was introduced into the Transformer model to reduce the model memory consumption further and the number of parameters and mitigate overfitting.The turbofan engine simulation dataset based on GasTurb modeling was used for verification.Compared with Transformer network and other traditional deep learning models,such as back propagation(BP)neural network,convolutional neural network(CNN),and recurrent neural network(RNN),the proposed model has improved prediction accuracy by 6.552%,28.117%,13.189%,and 10.29%,which proves its effectiveness,and it can provide a reference for aero-engine fault diagnosis.

关 键 词：航空发动机 故障诊断 自注意力机制 融合卷积Transformer 深度神经网络 

分 类 号：V263.6[航空宇航科学与技术—航空宇航制造工程]