基于时频谱图和粗糙集的柴油机故障图像纹理特征自动提取  

Automatic Extraction of Image Texture Feature of Diesel Engine Fault Based on CWT Time-Frequency Image and Variable Precision Rough Set Theory

作　　者：任金成[1] 肖云魁[1] 张玲玲[1,2] 沈虹[3] 封会娟[1] 

机构地区：[1]军事交通学院汽车工程系,天津300161 [2]军械工程学院火炮工程系,石家庄050003 [3]军事交通学院基础部,天津300161

出　　处：《内燃机工程》2015年第3期106-112,共7页Chinese Internal Combustion Engine Engineering

基　　金：总装备部预研资助项目(40407030302)

摘　　要：将图像纹理特征提取技术引入到柴油机连杆轴承磨损故障诊断中,首先采用连续小波变换对柴油机连杆轴承振动信号进行时频分析,为了减少循环波动的影响,将三个工作循环信号的时频分布平均为一个工作循环信号的时频图;然后将不同磨损工况的时频分布图转化为灰度图像,提取基于灰度共生矩阵四个角度的图像纹理特征参数;最后利用变精度粗糙集理论提取与故障程度强相关的特征参数。诊断实例表明:灰度共生矩阵能够反映柴油机连杆轴承不同磨损工况,变精度粗糙集可以从中提取出与故障程度强相关的五个关键参数用于分辨连杆轴承的四种磨损工况,小波时频图像特征提取和变精度粗糙集相结合能实现连杆轴承故障特征的自动提取。The image processing technique was introduced for the diagnosis of diesel engine conrod bearing wear fault. Firstly,vibration signal of engine conrod bearing was analyzed in time-frequency domain by the continuous wavelet transforms （CWT）. To eliminate the influence of engine cycle fluctuation, time- frequency analysis result of three-work-cycle signal was averaged to get the time-frequency of one work cycle signal. Then the time-frequency distributions of different abrade conditions were transformed into gray level images and the image texture features from four angles based on gray level co-occurrence matrix （GLCM） are extracted. Finally,variable precision rough set theory （VPRST） was used to extract the feature parameters that correlate with fault parts tightly. The experimental results indicate that different wear conditions of engine conrod bearing can be reflected by the gray level co-occurrence matrix,five key factors relevant to fault degree can be mined by VPRST for identifying four wear conditions of conrod bearing, and conrod bearing fault features can be extracted automatically by CWT time-frequency image combined with VPRST.

关 键 词：内燃机 连续小波变换 灰度共生矩阵 图像纹理特征 变精度粗糙集 特征提取 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]