Identification Of CO₂Welding Pool Penetration Characteristics Based On Vision

With the rapid development of manufacturing technology, CO2welding process alsorequires more and more high degree of automation, control of welding processparameters become a hotspot in research of the welding problem. Welding molten poolshape is closely connected with the process parameters of the welding process, to achievecontrol of the welding process parameters, through the morphological analysis ofwelding process parameters of molten pool, therefore, how to obtain the CO2weldingpool shape and identify the characteristics of the molten pool that become one of theimportant problems of the welding field. My master's thesis based on the visualcharacteristics of CO2welding depth of molten pool discriminant has carried on thethorough research.Under the short circuit transition welding in the process of CO2welding molten pool,the strong arc light and splash is still a research focus. Because of the influence ofvarious factors in the molten pool images will have splash and dust, etc., and makes theimage resolution reduced, how to effectively eliminate the noise also is the problemswhich need to solve urgently.First of all, This topic for collecting images will appear a series of problems, useingthe light source of the passive molten pool image visual sensor, select filming in themolten pool is looking behind the images of the molten pool, in the global shutter underthe way of molten pool images of high-speed photography, through specific treatments,the clear and complete welding molten pool images obtained, also for subsequent furtheranalysis of molten pool images provided favorable conditions.The second? This topic do some work of the welding molten pool images that wewere collected. First through pre-processing of image, which includes image geometrytransform, image enhancement and morphological image processing, etc., forpost-processing operation of image again, compare the gauss-Laplacian and Cannyoperator and other operators to extract the image edge, then we are using themathematical morphological method to image processing. Through a series of processing,to obtain the characteristic parameters of welding pool image information, and then theinformation to do further research.The last?This topic finally through the analysis of various parameters on themolten pool shape of the impact,and then set up a BP neural network model, neuralnetwork has the characteristics of highly nonlinear problem, through the network can beused to predict the penetration of the welding pool.Experiments show that the predicted data, comparing the data with the experimentalresults further proved that the neural network prediction enough under the condition ofdifferent welding process parameters of penetration degree, explaining the feasibility ofartificial neural network used in the field of welding.