声-固耦合模型下自适应高维特征焊缝缺陷超声检测

Ultrasonic Detection Method for Adaptive High-Dimensional Feature Weld Defects under Acoustic-Solid Coupling Model

针对焊缝缺陷种类识别精度有待提高的问题,提出了一种将随机森林与变分模态分解相结合的算法,以结合二者的自适应特征提取能力及对于高维特征的强适应特性。利用多物理场仿真软件建立含缺陷焊缝的声-固耦合模型,进行超声波无损检测仿真,得到含缺陷超声回波信号。利用变分模态分解求出其各IMF能量分布系数,并以其作为特征向量建立随机森林模型,进行焊缝缺陷识别。结果表明,利用该方法提取的缺陷回波信号特征能有效表征焊缝缺陷,以其建立随机森林模型可以对其进行准确识别。

Aiming at the problem that the recognition accuracy of weld defect types needs to be improved,an algorithm combining random forests and variational mode decomposition is proposed to make use of the adaptive feature extraction ability and the strong adaptive characteristics for high-dimensional features.The acoustic-solid coupling model with welding defects was established by multi-physics simulation software.Then ultrasonic non-destructive testing was simulated via these models to obtain three types of ultrasonic echo signals containing defects.These signals are decomposed by VMD method and the energy distribution coefficients of each IMF component are obtained.And the random forest model is established by these coefficients.The results show that the defect echo signal features extracted by this method can effectively characterize the weld defects,and the random forest model established with these features can be used to accurately identify them.