基于CNN电流数据形态识别的电动机故障诊断研究  

Research on motor fault diagnosis based on CNN current data morphology recognition

作　　者：龚敬群 李杰 黄冬明 周兴泽 马天雨 GONG Jingqun;LI Jie;HUANG Dongming;ZHOU Xingze;MA Tianyu(Baowu Equipment Intelligent Technology Co. ,Ltd. , Shanghai 201999, China;Hunan Normal University, Changsha 410083, Hu’nan, China;Hunan Ancun Technology Co. ,Ltd. , Changsha 410003, Hu’nan, China)

机构地区：[1]宝武装备智能科技有限公司,上海201999 [2]湖南师范大学,湖南长沙410083 [3]湖南安存科技有限公司,湖南长沙410003

出　　处：《宝钢技术》2021年第3期27-33,共7页Baosteel Technology

摘　　要：为实现层冷辊电动机由预防性批量更换向预测性定点维修转变的目标,提出一种基于卷积神经网络(CNN)电流数据形态识别的故障智能诊断方法。该方法首先对电流采集数据进行特征提取,并建立分类故障的各关键特征形态图形集;采用卷积神经网络识别故障下各特征的异常形态,建立以误报率和漏报率为目标的优化模型,通过以异常形态为基因片段的遗传算法寻优不同故障类型的各特征形态组合,建立分类故障形态组合模式库。现场数据验证结果表明所提方法能够满足现场漏报率和误报率要求。In order to achieve equipment management transformation from preventative batch maintenance to predictive fixed point maintenance for layered cold roll motors,an intelligent fault diagnosis method based on current CNN data morphology is proposed.Firstly,features are extracted from the current collected data,and an atlas of each key feature morphology is established for fault classification in this method.Using convolution neural network to identify all the features under the abnormal morphology with false alarm rate and false alarm rate optimization model is established with the target.With the use of abnormal morphology as a gene genetic algorithm to optimize combination of characteristics of different fault types of forms,and established the fault pattern classification combination pattern library.The field data verification results show that the proposed method can meet the requirements of missing alarm rate and false alarm rate.

关 键 词：CNN 形态识别 电动机故障诊断 大数据 人工智能 预测性运维 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]