【摘要】 多层导电结构涡流检测中,缺陷的自动识别和分类是急需解决的重要问题.提出了一种新的缺陷信号自动检测识别和分类方法,首先采用幅值中值预判和小波分析方法进行信号预处理,自动识别并提取包含缺陷的涡流检测信号片段;然后运用主分量分析法对含有缺陷的信号片段进行特征提取;接着构建最近均值、K近邻、BP网络和支持向量机四种分类器对缺陷信号进行分类;最后进行了实验研究,对多层导电结构三种形状缺陷的扫描检测信号进行识别和分类,验证了本文所提出方法的有效性,并比较了各分类器的性能,根据识别和分类错误率大小,可看出支持向量机分类器具有较好的鲁棒性和稳定性.更多还原

【Abstract】 In a number of industries, the automatic recognition and classification of defects in multi-layered structures are widely recognized as complex and urgent problems. This paper presents a novel method for automatic recognition and classification of defects during an eddy current (EC) inspection procedure. The signal segments containing possible defect events are detected based on computing the median of signals amplitude and the noise is eliminated using the wavelet packet analysis methods. The principal component analysis (PCA) is carried out to extract features from EC signals. The classification is performed using four different methods: nearest-mean classifier, k-nearest neighborhood classifier, the neural network and support vector machines. The method is tested on the eddy current signals from three different shapes of defect in the multi-layered structures. The results demonstrate the effectiveness of the proposed method. Compared with the classification accuracy as criteria, the best classifier recommended is support vector machines.更多还原