基于深度置信网络的制动器故障诊断方法  

Brake fault diagnosis method based on deep belief network

作　　者：李倩[1] 刘惠康[1] 皮瑶 喻青[2] Li Qian;Liu Huikang;Pi Yao;Yu Qing(Engineering Research Center of Metallurgical Automation and Measurement Technology of Ministry of Education,Wuhan University of Science and Technology,Wuhan 430081;Sinosteel Wuhan Safety and Environmental Protection Research Institute,Wuhan 430081)

机构地区：[1]武汉科技大学教育部冶金自动化与检测技术工程研究中心,武汉430081 [2]中钢集团武汉安全环保研究院,武汉430081

出　　处：《高技术通讯》2021年第10期1075-1080,共6页Chinese High Technology Letters

基　　金：国家重点研发计划(2017YFC0805100)资助项目。

摘　　要：针对吊车制动器故障诊断中故障机理复杂、先验知识欠缺、传统的故障诊断方法精度不高和人工依赖大等问题,本文提出一种基于深度置信网络(DBN)的制动器故障诊断方法。该方法通过柔性薄膜传感器阵列获取制动器不同工况的实时特征数据及信号,利用网络的无监督特征学习和有监督微调,构建制动器故障诊断的深层网络模型,从而实现了对制动器的故障诊断及预测。最后,分别与支持向量机(SVM)和遗传算法(GA)优化的BP神经网络(GA-BP)进行了对比研究,通过实验证明了本文方法的优越性。In order to solve the problems of complex fault mechanism,lack of prior knowledge,low precision and artifi-cial dependence of traditional brake fault diagnosis methods,a fault diagnosis method based on deep belief network(DBN)is proposed.In this method,the real-time characteristic data and signals of brake under different working conditions are obtained by flexible thin film sensor array,and the deep network model of brake fault is constructed by using unsupervised feature learning and supervised fine-tuning of training network,so as to realize the fault diag-nosis and prediction of the brake.Finally,the diagnosis results are compared with those of support vector machine(SVM)and BP neural network optimized by genetic algorithm(GA),which proved the superiority of this method by experiments.

关 键 词：深度置信网络(DBN) 吊车制动器 故障诊断 柔性薄膜传感器阵列 

分 类 号：TH17[机械工程—机械制造及自动化]