| Welding process is a complex process with multi-variables,nonlinearity and time delay.The modification of welding process parameters and assembly greatly affects the weld formation and joint quality.In the process of aluminum alloy welding,the traditional single sensing method cannot accurately and effectively monitor the welding process.Therefore,it is of great significance for the prediction and control of robotic TIG welding quality of aluminum alloy to use multi-source sensors to obtain welding assembly conditions,such as gap,mismatch,etc.,and to further predict the back penetration state combining with the front weld pool visual information.Based on the aerospace 2A14 aluminum alloy robotic TIG welding,a multi-source parameter acquisition system was built in this paper.Aiming at the weld pool image with strong arc interference and high dynamic range,the algorithm of strong arc filtering pretreatment and weld pool contour feature extraction was proposed.Based on the welding process parameters,assembly information and visual feature information of the front weld pool,a prediction model of the back weld width of 2A14 aluminum alloy TIG welding was established,a PID controller was designed,and the control verification experiment of variable heat dissipation "dumbbell" workpiece was carried out.Firstly,the hardware and software platform of multi-source information acquisition system for aluminum alloy TIG welding was built,including robot unit,welding equipment unit,vision sensing unit and parameter acquisition and control unit.The active vision sensing technology was used in the vision sensing unit,which can obtain stable and clear weld pool image.The parameter acquisition and control unit can realize highspeed acquisition of welding parameters and real-time output of the feedback control signal of current and wire feeding speed.Secondly,to deal with the processing of the high dynamic range weld pool image obtained in the process of aluminum alloy welding,a set of strong arc pretreatment and weld pool contour feature extraction algorithm was developed.A generative adversarial network was used to filter the arc light of the original weld pool image and generate a clear weld pool image without arc light interference.Furthermore,a contour extraction algorithm based on cascade regression tree was designed.The key points of the upper and lower edge of the weld pool were extracted and the contour was fitted.Finally,the geometric dimension feature value of the front weld width was obtained.With the help of the welding process parameters(welding current and wire feeding speed),assembly information(gap and mismatch)and front weld width information collected by the multi-source sensor information system,the prediction model of aluminum alloy TIG weld backside weld width based on Light GBM algorithm was constructed.The root mean square error of the predicted value was 0.68 mm.Compared with the traditional regression prediction algorithm SVR and ANN,the accuracy was increased by 32.6% and 45.4%.In order to solve the problem that the backside weld width was not synchronized with the front data,which was measured offline,the experimental system was designed to verify the correctness of the prediction model.Using the interpretability of the model,the characteristic importance of the parameters affecting the backside weld width was analyzed,and the factors affecting the penetration were analyzed in combination with the welding theory and production practice.Finally,based on the back weld width prediction model,a PID controller of TIG welding of aluminum alloy was designed with the back weld width as the controlled value and the welding current as the control value.The simulation and validation test of the controller were carried out.Under the interference of variable heat dissipation experimental conditions,the welding process keeps a good stability,and the back weld width was uniform.Compared with the constant standard experiment,the weld forming quality has been greatly improved. |
PDF Full Text Request