基于工况识别与自训练时空图卷积的航空发动机气路故障诊断  

Fault diagnosis of aero-engine gas path based on condition recognition and self-training ST-GCN model

作　　者：张世杰 胡家文 苗国磊 ZHANG Shijie;HU Jiawen;MIAO Guolei(School of Aeronautics and Astronautics,University of Electronic Science and Technology of China,Chengdu 611731,China;Chengdu Hangli(Group)Industrial Company Limited,Pengzhou 611936,China)

机构地区：[1]电子科技大学航空航天学院,四川成都611731 [2]成都航利(集团)实业有限公司,四川彭州611936

出　　处：《推进技术》2024年第11期245-254,共10页Journal of Propulsion Technology

基　　金：国家自然科学基金(72171037,71801168);四川省自然科学基金(2023NSFSC0476);四川省科技厅项目(2021ZDZX0004,2022ZDZX0036)。

摘　　要：航空发动机气路故障模式和运行工况多样,且相互耦合作用,使得同一故障模式在不同运行工况下表现特征存在差异性,增加了气路故障诊断的难度。提出一种基于多层感知器(MLP)工况识别与自训练时空图卷积网络(ST-GCN)模型的航空发动机气路故障诊断方法,利用高度、马赫数、燃油流量及高低压转子转速构建MLP对运行工况进行识别。利用发动机各截面状态监测参数构建图邻接矩阵,构建自训练ST-GCN半监督模型对相应工况下的故障模式进行诊断,采用自适应粒子群优化算法(APSO)对模型超参数进行寻优。采用燃气轮机仿真程序(GSP)生成发动机在动态工况下的状态监测数据,对提出方法的有效性进行验证。结果表明,先进行工况识别,再开展故障诊断,相较于忽略工况直接进行诊断,能够获得更高的故障诊断准确率,达到98.93%。The fault modes and operating conditions of an aero-engine gas path system are diverse and they interact with each other,resulting in variations in the characteristics manifestations of a same failure mode under different operating condition.This complexity increases the difficulty of gas path fault diagnosis.This paper pro⁃poses a fault diagnosis of aero-engine gas path based on multilayer perceptron(MLP)condition recognition and self-training spatial temporal graph convolutional neural network(ST-GCN)model.A MLP model was construct⁃ed to identify the operating condition based on altitude,Mach number,fuel flow,high-pressure rotor speed and low-pressure rotor speed.The cross-sectional state monitoring parameters of the engine were used to construct a graph adjacency matrix.A self-training ST-GCN semi-supervised model was developed based on monitoring pa⁃rameters of each cross-section state to diagnose the failure mode under the corresponding operating condition.An adaptive particle swarm optimization(APSO)algorithm was adopted for the selection of model hyperparameters.The effectiveness of the gas path fault diagnosis method was validated by the aero-engine state monitoring param⁃eters under dynamic operating conditions generated by gas turbine simulation program(GSP).The results indi⁃cate that,compared to directly diagnosing without considering operating conditions,the approach of first identify⁃ing the operating conditions and then conducting fault diagnosis can effectively improve the accuracy of fault diag⁃nosis.The fault diagnosis accuracy reaches 98.93%.

关 键 词：航空发动机 气路故障诊断 运行工况 自训练 燃气轮机仿真程序 

分 类 号：V263.6[航空宇航科学与技术—航空宇航制造工程]