Research Of Inspection Algorithms Of Corrugated Weld Seam Based On Vision

The current corrugated plate is still welded by human with inefficiency in the work.This thesis studies the detection algorithm of welds seam in corrugated plate based on vision.Combined with the actual needs of corrugated plate automatic tracking system,theoretical and engineering research is carried out,which has strong scientific and practical significance.For automatic welding of corrugated plates,the parts required to be welded are two pieces of protruding sheet metal.Because of the weld's special shape,laser cannot form laser lines on the corrugated plate.It is difficult to obtain effective measurement information.The thesis designs a seam detection and tracking system based on passive vision method.The micro camera is used to detect the information of weld's height difference,which provides a basis for controlling the angle of the welding torch.At the same time,the contact sensor is used to track the two-dimensional information,so that it can track the corrugated plate weld seam effectively.The most important segment of corrugated plate weld line detection is the image edge extraction algorithm.In traditional edge extraction methods,researchers always use artificial design features to extract edges.However,it has poor performance in using local information for the detecting of weld seam accuracy.At the same time,using a variety of features is too complex to be used.In this thesis,deep learning is introduced to detect weld seam,so as to overcome the defects of traditional methods.In the process of network communication,deep learning will add the activation layer in the neural network to make the network have stronger adaptability.In this paper,the exponential linear unit activation function is used to optimize the deep learning network.The improved network can converge better and faster than the previous network.This thesis uses Holistically-nested edge detection algorithm based on deep learning to automatically learn edge features.The algorithm learns not only local features,but also the whole picture's feature.It can greatly improve the accuracy of corrugated plate weld extraction.The Holistically-nested edge detection algorithm uses the fused image for output,with a defect of edge discontinuity.The thesis proposes the method of outputting feature maps on middle layer that has more obvious edge feature.It is proved to achieve better results.On the basis of theoretical research,this thesis carried out experimental tests.The thesis compared the improved Holistically-nested edge detection algorithm with the traditional algorithm.Experimental results show that the proposed algorithm in this thesis can accurately extract the edge of corrugated plate weld seam,which is obviously better than the traditional algorithm.The validity of the proposed algorithm is verified.