海军航空大学学报2026，Vol.41Issue(4)：653-662,10.DOI:10.7682/j.issn.2097-1427.2026.04.007
基于AC-CNN的航空发动机气路故障诊断算法
Aero-Engine Gas Path Fault Diagnosis Algorithm Based on AC-CNN
肖国松 1史宝舒 2江淼 3白杰2
作者信息
- 1. 中国民航大学科技创新研究院,天津 300300||中国民航大学民航航空器适航审定技术重点实验室,天津 300300
- 2. 中国民航大学安全科学与工程学院,天津 300300
- 3. 成都富凯飞机工程服务有限公司,四川 成都 610041
- 折叠
摘要
Abstract
To address the challenges of insufficient fault information extraction and low diagnostic accuracy in aero-engine gas path component fault diagnosis,an improved fault diagnosis model based on an AC-CNN is proposed,featur-ing a diagnostic mode of offline training and online diagnosis is proposed.First,the CBAM is optimized through chan-nel reduction and shared feature transformation,and embedded into a convolutional neural network.Then depthwise separable convolution is introduced to further reduce computational complexity,constructing an AC-CNN backbone for feature extraction.This design strengthens the model's feature representation capability and receptive field,thereby im-proving diagnostic accuracy.Finally,the refined high-quality features are fed into an ELM classifier to achieve fault classification.Experimental results show that the proposed model improves diagnostic accuracy to 99.29％while avoid-ing an excessive increase in computational complexity,requiring only 38.12％of the computational load of the tradi-tional CBAM module.The ELM classifier improves generalization and reduces overfitting.Compared to other models,AC-CNN shows superior robustness and is better suited for aero-engine gas path fault diagnosis.关键词
航空发动机/故障诊断/卷积神经网络/卷积块注意力模块/极限学习机Key words
aero-engine/fault diagnosis/convolutional neural network/convolution block attention mechanism/ex-treme learning machine分类
航空航天引用本文复制引用
肖国松,史宝舒,江淼,白杰..基于AC-CNN的航空发动机气路故障诊断算法[J].海军航空大学学报,2026,41(4):653-662,10.
