基于卷积神经网络的焊缝表面缺陷检测方法

针对工业激光焊接中,采用传统方法进行焊缝质量检测效率低下的问题,提出了一种基于卷积神经网络的工业钢板表面焊缝缺陷检测方法;首先基于卷积神经网络,搭建了一个多分类模型框架,并分析了各层中所用到的函数及相关参数;然后基于工业数控机床和工业相机进行了焊缝数据采集,并对这些数据进行了分类、增强、扩增等前期预处理;最后基于数控机器轴,采用滑动窗口检测的形式采集实际待测图像,并通过实验对比了传统的机器学习算法在该类图像数据中的性能评估;经实验证实,通过卷积神经网络训练得到的多分类模型,焊缝缺陷检测精度能达到97%以上,且每张待测图像的测试时间均在300 ms左右,远超机器学习算法,在准确性和实时性上均能达到实际工业要求。

Weld surface defect detection method based on convolution neural network

In order to solve the problem of low efficiency of traditional methods for weld quality detection in industrial laser welding, a method based on convolution neural network for detecting weld defects on the surface of industrial steel plate is proposed. Firstly, based on convolutional neural network, a multi classification model framework is built, and the functions and related parameters used in each layer are analyzed; Then, the weld data are collected based on Industrial CNC machine tools and industrial cameras, and these data are preprocessed by classification, enhancement and amplification; Finally, based on the CNC machine axis, the sliding window detection method is used to collect the actual image to be tested, and the performance evaluation of the traditional machine learning algorithm in this kind of image data is compared through experiments. The experimental results show that the accuracy of the multi classification model trained by convolution neural network can reach more than 97%, and the test time of each image to be tested is about 300ms, which is far beyond the machine learning algorithm, and can meet the actual industrial requirements in accuracy and real-time.