基于优化卷积神经网络的化工过程故障诊断方法  

Fault diagnosis method based on optimized convolutional neural network for chemical process

作　　者：胡宇鹏 郭丽杰[1] 张子龙 康建新[1,2] 崔超宇 乔桂英[1] HU Yupeng;GUO Lijie;ZHANG Zilong;KANG Jianxin;CUI Chaoyu;QIAO Guiying(School of Environmental and Chemical Engineering,Yanshan University,Qinhuangdao,Hebei 066004,China;Hebei Key Laboratory of Applied Chemistry,Yanshan University,Qinhuangdao,Hebei 066004,China)

机构地区：[1]燕山大学环境与化学工程学院,河北秦皇岛066004 [2]燕山大学河北省应用化学重点实验室,河北秦皇岛066004

出　　处：《燕山大学学报》2024年第6期550-560,共11页Journal of Yanshan University

基　　金：河北省自然科学基金资助项目(E2021203069);秦皇岛市科技计划项目(202101A319)。

摘　　要：为了从复杂化工过程的海量监测数据中提取出有效的故障特征,及时发现故障并准确识别故障原因,提出了一种基于优化卷积神经网络的化工过程故障诊断方法。首先,构建一维卷积神经网络二分类状态监测模型,以显著提高状态监测效率。其次,针对卷积神经网络无法评估网络特征数据重要程度的问题,引入注意力机制,通过为数据特征赋予不同权重,有效捕捉特征细节,抑制干扰信息,从而实现复杂化工系统中故障关键特征的自动提取,提高多种故障模式诊断的准确率。然后,针对卷积神经网络超参数优化手动设置建模效率低的问题,采用树型Parzen估计算法超参数优化技术,通过灵活的建模方式和高效的采集函数实现对超参数组合的自动精准调优,构建优化卷积神经网络的故障诊断模型。最后,采用田纳西-伊斯曼过程对所提出方法的有效性进行验证。结果表明,该方法能够及时、有效地监测并诊断出多种故障模式,可为维修人员提供可靠的决策依据。In order to extract effective fault features from the massive monitoring data of complex chemical process,a fault diagnosis method for chemical process based on optimized Convolutional Neural Network(CNN)is proposed to find faults in time and accurately identify the fault cause.Firstly,a one-dimensional CNN for binary classification is constructed to improve the efficiency of condition monitoring.Secondly,to address the issue that CNN cannot evaluate the importance of network:feature data,the attention mechanism is introduced into the fault diagnosis model of CNN,,which can ffectively capture feature details and suppress interference infornation.Different weights are assigned to the importance of the network feature data to realize the automatic extraction of key fault features for complex chemnical systems.Then,to tackle the low eficiency challenge in manual hyperparameter optimization,the Tree-structured Parzen Estimator(TPE)hyperparameter optimization technology is used to realize the precise tuning of byperparameter combination with flexible modeling method and efficient acquisition function.The optimized fault diagnosis model of CNN is constructed.Finally,the Tennessee Fastman(TE)process is used to verify the ffectiveness of the proposed method.The results show that the proposed method can detect multiple failure modes in a timely and efective manner,which can provide a reliable decision making basis for maintenance staff.

关 键 词：卷积神经网络 化工过程 故障诊断 注意力机制 超参数优化 

分 类 号：X937[环境科学与工程—安全科学]