基于卷积神经网络的激光超声缺陷检测研究

Research on laser ultrasonic defect detection based on convolutional neural network

针对超声缺陷检测受人为影响较大这一问题,提出一种基于卷积神经网络(CNN)的缺陷超声图像自动识别检测方法。利用激光超声可视化技术,对带有缺陷的金属部件进行检测得到最大振幅图像,并将其作为CNN缺陷检测模型的输入,CNN模型经过训练能很好实现对表面缺陷的自动检测。实验结果表明构建的CNN模型能够自主学习最大振幅图的缺陷特征并完成缺陷的自动识别,具有较高的检测效率,缺陷识别准确率可达97.22%,此外在缺陷连续检测中表现出较强的鲁棒性。

Recognizing that ultrasonic defect detection is greatly affected by humans,a method for automatic defect recognition and detection of ultrasonic images based on convolutional neural network(CNN)is proposed.In this paper,laser ultrasonic visualization technology is used to detect metal parts with defects to obtain the maximum amplitude image,and use it as the input of the CNN defect detection model.The CNN model can realize the automatic detection of surface defects after training.The experimental results show that the constructed CNN model can independently learn the defect characteristics of the maximum amplitude map and complete the automatic identification of defects.It has a high detection efficiency,and the defect recognition accuracy rate can reach 97.22%.In addition,it shows strong robustness in continuous defect detection.