基于灰度-梯度共生矩阵的旋转机械振动时频图形识别方法  

A fault diagnosis method based on gray level-gradient co-occurrence matrix of time-frequency image for rotating machinery

作　　者：窦唯[1] 刘占生[1] 

机构地区：[1]哈尔滨工业大学能源科学与工程学院,黑龙江哈尔滨150001

出　　处：《振动工程学报》2009年第1期85-91,共7页Journal of Vibration Engineering

基　　金：国家自然科学基金资助项目(50875056)

摘　　要：针对旋转机械振动图形信息在故障诊断中一直没有得到充分利用,在一定程度上影响诊断技术的推广和利用的问题。研究了基于灰度-梯度共生矩阵的旋转机械故障诊断方法,该方法利用能反映图形像素点灰度与梯度的分布规律的灰度-梯度共生矩阵,直接提取和挖掘旋转机械振动信号的时频灰度图形中的像素点与其邻域像素点空间关系的特征信息,有效地提取图形中纹理特征信息后,直接利用人工免疫反面选择算法实现旋转机械故障诊断。在600 MW模化汽轮机转子试验台上进行了转子正常、转子不平衡故障、转子不对中故障及轴承松动故障的试验,诊断结果表明可以获得较高的诊断精度,并与基于灰度共生矩阵的诊断方法进行了比较,证明该方法可以提高故障诊断的准确率,验证了该方法的可行性。Aims at the scarce use the vibration multi-dimensions image information of the rotating machinery in the fault diagnosis that affects the promotion and utilization of the diagnosis technology to some extent, a diagnosis method that is hased on gray level-gradient co-occurrence matrix is studied. This method directly extracts the dimensional relationship characteristics information between pixel points and their adjacent ones points in time-frequency gray level figures of the vibration signal for rotating machinery by using gray level-gradient co-occurrence matrix that can show the distribution law of gray level and gradient of the pixel points. Rotating machinery fault diagnosis can be directly conducted by using BP artificial neural networks after extracting the information of image texture characteristic. This method is validated to get high diagnosis accuracy by conducting the tests for normal rotor, unbalanced rotor, misaligned rotor and loose bearing pedestal on the modeling of 600 MW turbine experimental bench, and compared with the method based on gray co-occurrence matrix. The presented method is feasible and can improve the diagnosis accuracy.

关 键 词：旋转机械 故障诊断 共生矩阵 人工免疫 

分 类 号：TB752[一般工业技术—真空技术] TB53[理学—物理]