基于神经网络的超声导波锚杆剥离缺陷的识别

Identification of the Anchor Debonding by Ultrasonic Guided-wave Based on Neural Network

结合小波包分解技术与神经网络,提出了超声导波检测锚杆剥离缺陷的方法。首先,利用ANSYS-LSDYNA进行了数值模拟,在锚杆的顶端施加扫描激励波,得到了完好和剥离工况下的锚杆动态响应信号,利用小波包分解获得了缺陷参数与小波包能谱之间的关系,以小波包能谱作为输入向量训练适用于锚杆无损检测的神经网络,将剥离状况作为输出,建立了神经网络系统。最后利用混有不同噪声,且未参于训练的导波检测信号验证了神经网络检测的有效性。数值算例表明,该方法对噪声水平小于40%的检测信号,具有较好的识别结果。

Combining wavelet packet decomposition and neural network, the ultrasonic guided-wave was used to inspect the defects of the anchor debonding. To obtain the dynamic response signals of anchor bolt under integrated and debonding conditions, we calculated the anchor subjected to scanning excitation at the top by ANSYS-LSDY- NA. Using wavelet packet decomposition, we get the relationship between the parameters of defection and the wavelet packet energy spectrum. Considering the wavelet packet energy spectrum as input vector, while the debonding is considered as output vector, the neural network suitable for the anchor nondestructive testing was trained, and ultimately the neural network system was established. Furthermore, the tested pollution by noise was also used to identify the anchor integrity. The simulation showed an excellent identified results when noise level is less than 40%