基于传感技术的管道接头法兰环焊缝质量检测方法研究

Research on quality detection method of flange girth weld of pipe joint based on sensing technology

管道接头法兰焊缝位于管道内部,无法直接检测。为此提出基于机器视觉及卷积神经网络的识别方法。通过机器视觉传感器采集影像,采用张氏标定法和最小二乘积法去除镜头畸变,获得管道接头法兰图像。基于卷积神经网络训练管道接头法兰的特征,引入非线性激活函数令输出结果非线性化,通过池化运算降低特征复杂度和避免多度拟合情况发生,凭借Softmax将特征结果转化为分类概率,实现管道接头法兰环焊缝识别。实验证明,所提方法识别性能好,精度高,迭代周期短,误差均没有超过2mm,仅用100迭代就达到了最小识别误差值。

The flange weld is located inside the pipe,which cannot be directly detected.Therefore,a recognition method based on machine vision and convolutional neural network was proposed.The image was collected by ma⁃chine vision sensor,and the lens distortion was removed by Zhang’s calibration method and least square product method,and the image of pipe joint flange was obtained.Based on the feature of convolutional neural network trained the pipe joint flange,the nonlinear activation function was introduced to make the output result nonlinear,the feature complexity was reduced and the multi-degree fitting was avoided by pooling operation,and the feature re⁃sult was converted into classification probability with the help of Softmax to realize the identification of pipe joint flange girth weld.The experimental results showed that the proposed method had good recognition performance,high precision,short iteration period,and the error was less than 2 mm,and the minimum recognition error was reached with only 100 iterations.