基于小波分析和神经网络的传感器故障诊断  

Sensor Fault Diagnosis Based on Wavelet Analysis and Neural Network

作　　者：李子凡 叶志锋[1] 王彬[1] LI Zifan;YE Zhifeng;WANG Bin(College of Energy and Power Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 210016,China)

机构地区：[1]南京航空航天大学能源与动力学院,江苏南京210016

出　　处：《机械制造与自动化》2023年第4期196-201,共6页Machine Building & Automation

摘　　要：结合多维度特征提取和故障识别方法,提出一种基于小波分析和神经网络的传感器故障诊断方法。运用小波变换模极大值特征提取方法和高频小波能量特征提取方法,在小波分解层数和小波类型两个不同维度对传感器信号进行特征提取,提取的特征矩阵具有序列特性。研究结果表明:特征矩阵相对于特征向量,对不同信号具有更明显的区分度;运用LSTM神经网络对传感器进行故障诊断,根据不同压力工况下的传感器特征数据集,训练针对不同压力工况的LSTM神经网络预测模型,提高了预测模型的泛化能力;对LSTM神经网络预测方法进行试验验证,基于预测模型对随机压力工况下发生的随机故障进行预测,预测准确率达到98.33%。By combining multi-dimensional feature extraction and fault identification methods,a sensor fault diagnosis method based on wavelet analysis and neural network is proposed.Wavelet transform module maximum feature extraction method and high-frequency wavelet energy feature extraction method are applied to extract sensor signal features in two different dimensions of wavelet decomposition level and wavelet type,and the extracted feature matrix has sequence characteristics.The research results show that the feature matrix has a more distinct degree of discrimination for different signals than the feature vector.The LSTM neural network is used to diagnose the fault of the sensor.According to the sensor feature data set under different pressure conditions,the LSTM neural network prediction model for different pressure conditions is trained,which improves the generalization ability of the prediction model.The LSTM neural network prediction method is tested and verified.Based on the prediction model,the random failures occuring under random stress conditions are predicted,and the prediction accuracy rate reaches 98.33%.

关 键 词：故障诊断 小波多分辨率分析 多维度特征提取 LSTM神经网络 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]