【摘要】 活塞作为发动机内最重要的零件之一，工作过程中将承受巨大的爆发力。当活塞喉口存在微细缺陷时，爆发力将会导致缺陷开裂从而产生严重的安全隐患，因此对于活塞喉口微细缺陷检测的研究具有重要意义。采集三种不同缺陷类型的活塞涡流信号，进行降噪后提取其时域、频域及时频域内多重特征^[1]。对比分析了基于主成分分析（Principal Component Analysis,PCA）和线性判别分析（Linear Discriminant Analysis,LDA）降维的活塞喉口微细缺陷检测与识别方法，并对降维结果分别进行线性判别分类和高斯朴素贝叶斯分类（Gaussian Naive Bayes,GaussianNB），对比缺陷识别的准确率与模型训练时间，从而得出性能最好的缺陷识别模型。实验结果证明LDA-GaussianNB模型可高效判别活塞喉口微细缺陷类型。更多还原

【Abstract】 As one of the most important parts in the engine,the piston will bear huge explosive force in the process of work. When the piston throat has a micro defect,the explosive force will lead to defect cracking and endanger the safety of the vehicle. Therefore,the research on the micro defect detection of the piston throat is of great significance. Three kinds of piston eddy current signals with different defect types are collected and their multiple characteristics in time domain,frequency domain and frequency domain are extracted after noise reduction. Based on Principal Component Analysis(PCA)and Linear Discriminant Analysis(LDA),dimension reduction methods for detecting and identifying fine defects in piston throat were compared,and the dimension reduction results respectively Linear Discriminant classification and Gaussian Naive Bayes classification,contrast defect recognition accuracy and the model of training time,thus it is concluded that the best performance defect recognition model. The experimental results show that the LDA-GaussianNB model can effectively identify the defect types of piston throat.更多还原