基于小波和神经网络的焊接缺陷识别方法

Recognition Method of Weld Flaws Based on Wavelet and Neural Network

超声检测是钢结构焊缝质量检验的常用方法,本文针对超声检验中缺陷类型识别困难的问题,提出一种定性方法.首先利用小波包变换提取超声回波信号的能量作为特征向量,然后将得到的特征向量输入到BP神经网络中,应用于裂纹、气孔、未焊透三类缺陷,识别率达到了86.7%.结果表明:基于小波包变换和BP神经网络的钢结构焊缝缺陷定性方法是十分有效的.

Ultrasonic testing is the common method to inspect the weld quality of steel construction. But it is difficult to identify the type of flaws. In light of this problem, this paper proposes a recognition method. Firstly, we use the energy from extracting the ultrasonic echo signal when wavelet packet transform as feature vector, and then put it into a BP neural network for classifying the crack, gas pore and incomplete weld flaws, the recognition rate reaches 86.7%. The results show that the recognition method proposed for steel construction weld flaws based on wavelet packet transform and BP neural network is extremely effective.