改进注意力机制的航空发动机试验转子系统智能故障诊断  

Improved attention mechanism for intelligent fault diagnosis ofexperimental rotor systems in aero engines

作　　者：伍济钢 文港 杨康 WU Jigang;WEN Gang;YANG Kang(Hunan Province Key Laboratory of Health Maintenance Equipment,Hunan University of Science and Technology,Xiangtan 411201,China)

机构地区：[1]湖南科技大学机械设备健康维护湖南省重点实验室,湖南湘潭411201

出　　处：《振动与冲击》2024年第4期261-269,共9页Journal of Vibration and Shock

基　　金：国家自然科学基金(51775181)。

摘　　要：考虑到航空发动机的工作环境十分恶劣,其故障的振动信号特征隐蔽且噪声干扰严重,为了加强网络对振动信号中关键特征的提取能力,提出了改进注意力机制的航空发动机转子系统智能故障诊断方法对航空发动机转子系统的不平衡和碰摩等故障进行诊断。提出局部池化改进的通道注意力机制,能够通过预提取局部极值解决现有通道注意力机制对航空发动机转子故障通道信息提取能力不足的问题;提出多评分机制改进的空间注意力机制,能够通过不同尺度的卷积评分解决现有空间注意力机制对航空发动机转子故障空间信息提取能力不足的问题;将二者结合构建改进的通道空间注意力机制模块,再导入一维卷积神经网络中构建改进注意力机制的一维卷积神经网络完成智能故障诊断,并且通过航空发动机转子系统故障数据集对比分析试验证明了该网络优秀的检测性能、抗噪性能和泛化性能等综合性能以及注意力机制改进方法的可行性。Considering the harsh working environment of aero-engine,the hidden characteristics of the vibration signal of the fault and the serious noise interference,in order to strengthen the extraction ability of the network to the key features in the vibration signal,an intelligent fault diagnosis method for aero-engine rotor system with improved attention mechanism was proposed to diagnose the faults of an aero-engine rotor system such as unbalance and friction.The proposed improved channel attention mechanism with partial pooling can solve the problem of insufficient information extraction capability of the existing channel attention mechanism for aero-engine rotor fault channels by pre-extracting local extreme values.The improved spatial attention mechanism with multiple scoring mechanism was proposed,which can solve the problem that the existing channel attention mechanism has insufficient ability to extract spatial information about aero-engine rotor faults by convolution scoring of different scales.The above methods were combined to build an improved attention mechanism module in the channel space,and then imported into a 1D convolutional neural network to build a 1D convolutional neural network with improved attention mechanism to complete intelligent fault diagnosis,and the comprehensive performance of the network such as excellent detection performance,noise immunity and generalization performance as well as the feasibility of the attention mechanism improvement method were demonstrated by the comparative analysis of the aero-engine rotor system fault dataset.

关 键 词：航空发动机转子 故障诊断 注意力机制 卷积神经网络 

分 类 号：TV232[水利工程] U226.81[交通运输工程—道路与铁道工程]