CAJ | 학술논문

提出了一种基于核主元分析（KPCA）和遗传BP神经网络的滚珠丝杠故障诊断方法。首先用2个测点的6个传感器同步采集滚珠丝杠的振动信号，并进行特征提取，得到原始样本空间，然后利用核主元分析对原始样本空间进行降维处理，以消除样本间的冗余信息。引入遗传算法，解决了传统BP神经网络初始权值和阈值选择的随机性，并建立3种不同的滚珠丝杠故障诊断网络对滚珠丝杠的正常状态、丝杠弯曲、滚珠破损和滚道磨损4种状态进行诊断实验。结果表明：基于核主元分析和遗传BP神经网络的滚珠丝杠故障诊断方法明显地缩短了网络的训练时间，有效地提高了故障状态的识别率。
제출료일충기우핵주원분석（KPCA）화유전BP신경망락적곤주사강고장진단방법。수선용2개측점적6개전감기동보채집곤주사강적진동신호，병진행특정제취，득도원시양본공간，연후이용핵주원분석대원시양본공간진행강유처리，이소제양본간적용여신식。인입유전산법，해결료전통BP신경망락초시권치화역치선택적수궤성，병건립3충불동적곤주사강고장진단망락대곤주사강적정상상태、사강만곡、곤주파손화곤도마손4충상태진행진단실험。결과표명：기우핵주원분석화유전BP신경망락적곤주사강고장진단방법명현지축단료망락적훈련시간，유효지제고료고장상태적식별솔。
A fault diagnosis method of Ball screw based on KPCA and Genetic GA-BP neural networks was proposed. First,syn-chronous acquisition of vibration signal of the Ball screw using 6 sensors in 2 points was done,and the original sample space was ob-tained by feature extraction. Then the dimension of the original sample space was reduced with the KPCA to eliminate the redundant in-formation of the sample space. By introduced Genetic Algorithm,the randomness at selecting of traditional BP neural network initial weights and threshold was resolved,and three network in different types were established to diagnosis four different state of Ball screw including normal state,screw bending,broken ball and raceway wear. Results show that,ball screw fault diagnosis method based on KPCA and GA-BP neural network has significantly shorten the training time of the network,and effectively improve the recognition rate of the fault condition.