基于深度信念网络的铁谱图像智能识别方法与试验验证  

Intelligent Recognition Method and Experimental Verification of Ferrography Image Based on Deep Belief Network

作　　者：樊红卫[1,2] 胡德顺 高烁琪 张旭辉 曹现刚[1,2] FAN Hongwei;HU Deshun;GAO Shuoqi;ZHANG Xuhui;CAO Xiangang(School of Mechanical Engineering,Xi'an University of Science and Technology,Xi'an Shaanxi 710054,China;Shaanxi Key Laboratory of Mine Electromechanical Equipment Intelligent Monitoring,Xi'an Shaanxi 710054,China)

机构地区：[1]西安科技大学机械工程学院,陕西西安710054 [2]陕西省矿山机电装备智能监测重点实验室,陕西西安710054

出　　处：《润滑与密封》2021年第7期15-22,共8页Lubrication Engineering

基　　金：国家自然科学基金项目(51834006);陕西省自然科学基础研究计划项目(2021JM-391)。

摘　　要：为了提高机械设备磨损状态识别精度和效率,利用深度学习中的深度信念网络并结合铁谱分析技术,提出一种磨粒图像智能识别方法。首先,建立受限玻尔兹曼机模型,将其用于深度信念网络的预训练,初始化网络模型参数,完成识别模型构建;然后,利用铁谱分析技术,通过铁谱仪制备铁谱图像,进行图像预处理,得到学习样本;最后,对网络模型参数进行研究,观测各参数变化对模型性能的影响规律,得到最优取值。结果表明:所提出的方法能够快速准确地识别设备磨损类型,识别正确率达到99%以上。In order to improve the accuracy and efficiency of wear condition recognition of mechanical equipment,an intelligent recognition method of wear particle images by using Deep Belief Network(DBN)model in deep learning with ferrograph analysis technology was proposed.The Restricted Boltzmann Machine(RBM)was built,which was used to pretrain the DBN,and initialize the parameters of DBN,so as to complete the construction of identification model.By using ferrography analysis technology,the ferrography image was prepared by a ferrography instrument and preprocessed to build the sample used for DBN.The parameters in DBN were studied,and the influence of the changes of every parameter on the model performance was observed to obtain the optimal values.The results show that the proposed method can quickly and accurately identify the wear types of equipment,and the recognition accuracy reaches more than 99%.

关 键 词：故障诊断 铁谱分析 图像识别 深度学习 深度信念网络 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TH17[自动化与计算机技术—计算机科学与技术]