构件裂纹缺陷的超声识别

Ultrasonic recognition of crack defect of component

将小波包多分辨率分析与能量谱相结合,提出了金属材料缺陷特征提取的方法(小波包子带能量比较法)及不同缺陷的识别方法(小波分形神经网络法)。选取最能反映缺陷特征的参数——"能量特征向量"作为特征参数,进行缺陷的特征提取。缺陷的识别方法将小波包分解后各子带系数的分形维数作为特征矢量,对其进行径向基神经网络训练,从而可很明显区分出有无裂纹以及不同裂纹信号。以航天发射塔架钢连接构件疲劳裂纹超声检测信号为例,使用所提出的特征提取和模式识别方法,结果表明是行之有效的新方法,为金属材料缺陷检测与识别开拓了新思路。

Through combination of multi-resolution signal decomposition of the wavelet packet and energy spectrum, the extraction method of defect features of metal matrix composite （ multi-resolution analysis of warelet packet） and recognition method of different defects （wavelet-fractal-NN） were put forward. The energy characteristic vectors which can reflect the defect features were chosen as characteristic parameters for defect recognition, The fractal dimensions of sub-bands from wavelet packet were used as characteristic vectors during the defect recognition, then the vectors were trained by RBF NN so as to identify different crack signals and non-crack signals. Taking ultrasonic detection signal of steel component of launching tower as example, the feature extraction and recognition methods were effective, which can provide a new idea for detection and recognition of the defect in composite material.