Research On Welding Seam Identification Technology In Armored Car Body Welding Process

The welding of armored vehicle bodies is usually carried out using manual welding techniques that rely on real-time observation by the welder to adjust the welding position and welding distance,resulting in low productivity and making it difficult to meet the welding requirements of modern production.To solve this problem,intelligent welding technologies such as robotic welding are used,where laser vision sensing technology is used to project a laser onto the surface of the welded workpiece and detect the weld image containing the weld profile information to obtain the weld characteristic parameters,which has been very widely used in weld identification due to its non-contact and high accuracy characteristics.In the actual welding process,the captured weld image will undoubtedly be disturbed by noise such as spatter,fume and arc light,which affects the quality and efficiency of the welding and does not guarantee the stability and accuracy of the welding.In order to improve the adaptability of weld seam feature extraction and overcome the influence of the external environment,it is important to study the learning of adaptable weld seam feature extraction techniques.To this end,this thesis uses a deep learning method to carry out research on weld seam feature recognition technology,which can effectively achieve the accurate extraction of weld seam features.Firstly,the working principle of a weld seam feature extraction system consisting of three models: the camera model,the structured light plane model and the hand-eye model is investigated.The camera model is used to acquire images of the weld seam,the structured light plane model is used to extract geometric information about the surface of the weld seam and the hand-eye model is used to describe the relative position of the camera and the weld seam.In order to realize the calibration of these models,methods for the calibration of the internal and external parameters of the camera,the calibration of the structural light plane parameters and the calibration of the hand-eye matrix parameters are investigated.Parameter optimization methods are also investigated to further improve the performance and efficiency of the system.Then,the features of the weld image are analyzed in depth,and the problem of feature point identification in the weld image is transformed into a key point detection problem.Using the advantages of convolutional neural networks in feature extraction,a key point detection method for weld feature point extraction is proposed.Through the weld seam feature extraction network,a heat map of the weld seam feature points is output,where the location with the largest response value in the heat map is the key point.By obtaining the coordinates of the key points,the identification and localization of the weld seam is achieved.Finally,the design and experiments of the weld seam image feature extraction system are completed.Based on the triangulation principle,the optimal method is selected and the component parts of the weld seam feature extraction system are selected.The weld seam features of multiple bevel types are extracted through experiments and the results of the experiments are analyzed.The experimental results show that the root mean square error of weld feature point positioning for the method used in this thesis is 0.187 mm.The network model designed in this thesis has high detection accuracy in the weld feature point identification task and is highly adaptable and generalizable to meet the requirements of automatic welding.