摘要
用锥形核时频分布对柴油机气阀机构8种状态下的缸盖表面振动信号进行了时频分析,将时频分析结果用灰度图像显示出来。对时频图像进行归一化处理,然后采用支持向量机直接对归一化后的图像进行分类,从而将气阀机构的故障诊断转换为对时频图像的识别。试验结果表明,支持向量机不需要先对时频图像进行特征提取就可以取得比较好的分类效果。将多类分类问题分解为多个两类分类问题时,根据整体的分类效果对支持向量机进行参数优化可以得到更高的分类正确率。
The Cone-Shaped Kernel Distributions of eight kinds of vibration acceleration signals, acquiring from the cylinder head in eight different states of valve train, were calculated and expressed in grey images and a series of time-frequency images were obtained. Support Vector Machines (SVMs) were directly (adopted) to classify the normalized images. In this way, the process of fault diagnosis for valve train was shifted to the classification of time-frequency images. The experimental results showed that a high recognition rate could be obtained by the method of time-frequency analysis and SVM. Using SVM, since there was no need to extract features from time-frequency images before classifying, the fault diagnosis process could be simplified. The results also showed that for multi-class problems, optimization of the parameters of SVM over all 2-calss SVMs could obtain a more satisfied classifying accuracy compared to the optimization of parameters on every 2-calss SVM.
出处
《内燃机学报》
EI
CAS
CSCD
北大核心
2004年第3期245-251,共7页
Transactions of Csice
基金
863"计划资助项目(2001AA411310)
国家自然科学基金资助项目(50375115)。
关键词
柴油机
故障诊断
时频分析
支持向量机
气阀故障
Diesel engine
Fault diagnosis
Time-frequency analysis
Support vector machine
