Research On Seam Tracking System Based On Laser Vision Guidance

While the welding manufacturing industry in China is developing towards the direction of high precision,high efficiency and low energy consumption,there are also some problems that need to be solved urgently.For example,under the influence of complex and changeable welding environment,it is still difficult to identify welds;When the position of the weldment is shifted or deformed due to the influence of high temperature of the welding gun,the error between the actual welding track and the preset track will be caused.This paper focuses on the optimization and innovation of weld image recognition and weld trajectory planning.After the calibration of each part of the welding system,the convolution neural network is used to detect and identify the weld seam,and a series of processing is carried out on the collected images.On this basis,an optimization method based on ant algorithm is proposed to realize weld trajectory planning and automatic welding control.The specific work is as follows:(1)Analyze the imaging principle of the camera,establish the mathematical model of the camera,and use MATLAB software to calibrate the parameters inside the camera according to the zhang’s calibration principle,and obtain the parameters.Establish an experimental platform based on hand-eye model,obtain the relative position between the camera and the manipulator,calculate the incidence matrix,and complete the conversion between camera coordinates and spatial coordinates.(2)In order to improve the accuracy of weld detection and recognition in welding operations,a series of image processing technologies such as image filtering and morphological changes are carried out on laser welding images in this paper.Through performance testing and experimental analysis of various algorithms,the noise reduction effect of different methods is compared,and the edge detection technology of Sobel operator,Canny operator and Roberts operator is used to realize the extraction of laser stripe centerline.(3)A weld seam feature extraction model based on improved Faster RCNN is proposed.Res Net network is used to replace VGG network,which optimizes the network parameters and structure,overcomes the problems of low detection accuracy,high time complexity and vulnerability to environmental impact of traditional models,and reduces the gradient disappearance problem caused by increasing depth in depth neural network.In the research of FPN algorithm,firstly select a small number of samples for training,and then expand the samples to achieve high accuracy.The experimental results show that the confidence of most detection results is above 98%,and the accuracy rate is 99.6%.(4)The traditional ant colony algorithm is improved and applied to the weld path planning.The artificial potential field is introduced to enhance the pertinence of the ant colony to the target search.The redundant polylines in the path are cut according to the path optimization pruning strategy,and finally the B-spline curve strategy is used to smooth the path.Experiments show that this method can effectively reduce the average and shortest distance of the welding path,and can quickly find the optimal path.The system is transmitted to ABB’s execution terminal via Ethernet to realize intelligent welding.