基于PCA-LSTM的轧制过程故障诊断  

Rolling Process Failure Diagnosis Based on PCA-LSTM Neural Network

作　　者：张瑞成 许阳 梁卫征 ZHANG Rui-cheng;XU Yang;LIANG Wei-zheng(North China University of Technology,College of Electrical Engineering,Hebei Tangshan 063210,China)

机构地区：[1]华北理工大学电气工程学院,河北唐山063210

出　　处：《机械设计与制造》2024年第10期226-229,共4页Machinery Design & Manufacture

基　　金：河北省自然科学基金(F2018209201)。

摘　　要：针对精轧部分辊缝控制执行器失效和冷却水阀故障难以诊断,且轧制过程具有数据结构复杂、前后时刻有影响、非线性强等特性,提出了基于PCA-LSTM的轧制故障诊断模型。使用主成分分析(PCA)提取精轧部分关键特征,提高了轧制过程网络训练时的迭代速度,降低了LSTM网络的输入维度和预测难度。将提取后的特征数据作为LSTM网络的输入,将故障类别作为输出,使用PCA-LSTM故障诊断模型对两类故障进行诊断。通过比较PCA-LSTM、LSRM、BP神经网络诊断结果表明,采用PCA-LSTM诊断模型后的数据训练网络迭代速度更快,且诊断正确率超过99%,诊断效果优于普通LSTM和BP神经网络。Aiming on the failure and the failure of cooling water valve,the complex data structure,influence and nonlinear characteristics.The extraction of the main component analysis(PCA)improves the iteration speed during the rolling process network training and reduces the input dimension and prediction difficulty of the LSTM network.Use the extracted feature data as the input to the LSTM network and the fault category as the output.Diagnosis of two types of faults using the PCA-LSTM troubleshooting model.Comparing the diagnostic results of PCA-LSTM、LSRM、BP neural networks shows that the data training network iteration after using the PCA-LSTM diagnostic model is faster and the diagnostic accuracy exceeds 99%,with better diagnostic effect over ordinary LSTM and BP neural networks.

关 键 词：LSTM 主成分分析 轧制过程 故障诊断 

分 类 号：TH16[机械工程—机械制造及自动化] TP183[自动化与计算机技术—控制理论与控制工程]