内燃机振动时频图像的编码特征提取与诊断
Coding Feature Extraction and Diagnosis of I.C. Engine Vibration Time-frequency Images

针对传统内燃机振动诊断方法在参数选择和特征提取方面的难题，提出一种将S变换和模块二维主成分分析（M-2DPCA）相结合的内燃机故障诊断方法。该方法首先利用S变换将采集到的内燃机缸盖表面振动信号生成振动谱图像；然后通过M-2DPCA对图像矩阵进行模块化处理，利用所有样本子图像构建总体散布矩阵，计算最优投影向量，进行图像特征参数提取；最后，利用最近邻分类器进行分类识别，完成诊断。将该方法应用于内燃机气阀机构8种工况下振动信号的诊断实例中，识别率可达到94.17%，证明了该方法的有效性。
According to the problems of parameter selection and feature extraction for vibration diagnosis of traditional internal combustion (I.C) engine, a new fault diagnosis method is discussed. The method based on S-transformation and Module Two Dimensional Principal Components Analysis (M-2DPCA) is proposed to carry out fault diagnosis of I.C. engine valve mechanism. First of all, the method transfers cylinder surface vibration signals of I.C. engine into images through S-transform. Second extracting the optimized projection vectors from the general distribution G which is obtained by all sample sub-images, so that vibration spectrum images can be modularized using M-2DPCA. At last, these features matrix obtained from images project will served as the enters of nearest neighbor classifier, it is used to achieve fault types' division. The method is applied to the diagnosis example of the vibration signal of the valve mechanism eight operating modes, recognition rate up to 94.17%; the effectiveness of the proposed method is proved