基于SVM及电流牵扯效应的金属缺陷分类识别方法

金属在服役期间经常存在一些由应力、腐蚀和疲劳造成的缺陷，在众多的缺陷检测技术中，交流电位法作为一种无损检测技术在检测腐蚀坑和裂纹方面得到了广泛应用。用交流电位法检测不同缺陷时，由于几何形态的差异，缺陷深度的计算方法也不相同。因此需要在计算缺陷深度之前对所检测区域的缺陷类型做出识别。本文目的在于寻找一种高精度的缺陷分类识别方法。并且针对腐蚀坑和裂纹这两种最常见的金属缺陷，根据其对电流的牵扯效应不同，提出利用邻近检测区域的4个牵扯因子作为缺陷区域的特征向量建立分类模型。在大量仿真计算的基础上，分别建立坑蚀和裂纹的特征向量集，并由这些数据集训练得到基于遗传算法（GA）优化的支持向量机（SVM）分类模型。仿真测试结果中数据测试集分类精度较高，平板实验也得到了较高的识别精度。实验结果表明文中提出的缺陷分类识别方法对腐蚀坑和裂纹的分类识别具有很高的精度。

Method for defect classification based on SVM and current drag effect

Metal structures often process defects caused by stress, corrosion and fatigue. Among kinds of non-destructive testing technologies, alternating current potential drop technique (ACPD) as a valid method has been widely used in pitting corrosion and crack measurement. When ACPD is used to test different kind of defects, due to the various geometry shapes, the approach to solve defect depth is different as well. Therefore, it is essential to classify and identify the measured defects in the test region. The purpose of this paper was to find an approach which can classify different defects accurately. In this paper, we studied on two kinds of typical defect based on an eigenvector which consists of 4 adjacent drag factors. After large quantities of simulation, the pitting and crack eigenvector datasets were built. Moreover, a support vector machine (SVM) optimized by genetic algorithm and trained by eigenvector datasets was obtained. Simulation test data showed that the trained and optimized SVM model has a high classification accuracy, and the metal plate experiment also indicated that the model has a good precision in actual defect classification. All the experiments showed that the proposed approach used for defect classification had high precision in pitting and crack classification.