基于优化堆叠自编码器的故障诊断方法  

Fault Diagnosis Method Based on Optimized Stacked Autoencoder

作　　者：李启泽 徐琛[1,2] 陶洪峰 杨慧中[1,2] LI Qize;XU Chen;TAO Hongfeng;YANG Huizhong(School of Internet of Things Engineering,Ministry of Education,Jiangnan University,Wuxi 214122,China;Key Laboratory of Advanced Process Control for Light Industry,Ministry of Education,Jiangnan University,Wuxi 214122,China)

机构地区：[1]江南大学物联网工程学院,江苏无锡214122 [2]江南大学教育部轻工过程先进控制重点实验室,江苏无锡214122

出　　处：《控制工程》2024年第2期231-237,共7页Control Engineering of China

基　　金：国家自然科学基金资助项目(62103167,61773181,61203092);中央高校基本科研业务费专项资金资助项目(JUSRP51733B);江苏省自然科学基金资助项目(BK20210451)。

摘　　要：针对当前堆叠自编码器故障诊断方法在识别有效分类特征方面的缺陷,提出一种基于Fisher判别准则优化的堆叠自编码器故障诊断方法。将Fisher判别准则中寻找最佳投影方向的特征学习方法融入到堆叠自编码器的预训练中,利用样本标签信息在堆叠自编码器的逐层非线性映射中学习最佳的投影方向。在Fisher判别准则优化的损失函数约束下训练,增加不同类别故障特征的类间距离,减小同类别特征的类内距离。由于在堆叠自编码器的预训练中,同时设计最小化重构特征和最大化分类特征的约束条件,预训练后的堆叠自编码器能够提取到更有效的特征信息,以提升最终故障诊断的准确率。通过在Tennessee Eastman(TE)化工过程的应用验证了所提故障诊断方法的可行性。For the shortcomings of current stacked autoencoder fault diagnosis methods in identifying effective classification features,a stacked autoencoder fault diagnosis method optimized based on Fisher discriminant criterion is proposed.The main idea is to apply the feature learning method of finding the optimal projection direction in Fisher discriminant criterion to the pre-training of stacked autoencoder,and the optimal projection direction is learned from the layer-by-layer nonlinear mapping of stacked autoencoder by using labeled sample information.The training process is performed under the constraint of loss function improved by Fisher discriminant criterion,and the distance between different fault types are increased,while the distance between the same fault types are reduced.Since the pre-training of the stacked autoencoder design the constraints of minimizing the reconstruction features and maximizing the classification features,the pre-trained stacked autoencoder can extract more effective feature information and improve the accuracy of the final fault diagnosis rate.Through the simulation of Tennessee Eastman process,the feasibility of the proposed fault diagnosis method is verified.

关 键 词：堆叠自编码器 FISHER判别准则 故障诊断 特征提取 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]