应用SIFT_BoW和HIK_SVM的行星轮齿面磨损故障诊断  

Wear-out Failure Diagnosis of Planetary Gear Tooth Surface based on SIFT_BoW and HIK_SVM

作　　者：任川 崔宝珍[1] 王斌 Ren Chuan;Cui Baozhen;Wang Bin(College of Mechanical Engineering,North University of China,Taiyuan 030051,China)

机构地区：[1]中北大学机械工程学院,太原030051

出　　处：《机械科学与技术》2020年第7期1000-1004,共5页Mechanical Science and Technology for Aerospace Engineering

基　　金：山西省自然科学基金项目(2009011026-1);山西省研究生创新基金项目(2008072)资助。

摘　　要：行星轮齿面磨损故障信号具有特征薄弱、特征量少等缺点,对其进行故障特征识别较为困难。本文中提出一种新的方法:首先,将原始振动加速度信号幅值作为像素点构造灰度图像,检测灰度图像的特征点并对检测出的特征点向量描述;然后将灰度图像的特征描述向量聚类,构建词袋模型;最后用直方图相交核支持向量机算法对其进行分类。该方法不但不需要对原始信号模态分解和降噪处理,还可以提取出大量的信号特征,提高了故障特征识别的效率和准确率。对正常轮齿、2个齿面磨损和3个齿面磨损故障进行了诊断实验,准确率高达98.55%,实验结果验证了所提方法的有效性。The vibration signal of planetary gear tooth surface wear-out failure has some disadvantages such as weak feature and little feature quantity.Therefore,it is difficult to identify the wear-out failure feature of planetary gear tooth surface.In order to solve this problem,a new fault diagnosis method is put forward.Firstly,the amplitude of the original vibration acceleration signal is used as pixels to construct a gray-scale image,and the feature points of the gray-scale image are detected and the detected feature point vectors are described.Then,the feature description vectors of the gray-scale image are clustered to construct a bag of words model.Finally,the histogram intersection kernel support vector machine algorithm is used to classify the gray-scale image.This method not only does not need to decompose the original signal and reduce the signal noise,but also can extract a large number of signal features,and improve the efficiency and accuracy of fault feature detection.Experiments were carried out to diagnose the normal gear teeth,fault gears with two tooth surfaces wear and three tooth surfaces wear,and the diagnosis accuracy was up to 98.55%.Experimental results show that the proposed method is effective.

关 键 词：行星轮 齿面磨损 特征点 词袋 支持向量机 

分 类 号：TH17[机械工程—机械制造及自动化]