Penetration State Recognition And Predictive Control Of GTAW Process Based On Multi-information Fusion

In the process of GTAW welding,the effective detection and control of weld penetration state have always been the difficulty and focus in the field of automatic welding research.The development of sensor technology,the research and application of intelligent and automatic control theory methods,these methods and researches provide an important idea to solve this problem.In the past,GTAW welding process fusion state detection and control involved only one sensor mostly,or it involved in multiple-sensor fusion to detecte GTAW weld penetration status but no come to form a closed loop control,and it didn’t add the infrared thermal image information into the GTAW sensing information mining system,Therefore,there are certain limitations in the previous studies,resulting in the accuracy of penetration state detection results and the reliability of control have certain deficiencies.For GTAW welding process,this article is based on visual sensor,infrared thermal image sensor,sound sensor and other sensors to monitor the welding arc weld penetration in the process of state information,study of GTAW welding process feature extraction method of the multi-sensor information extracted from the original information were collected to each sensor characteristic information about welding penetration.Furthermore,the fusion method of multi-sensor characteristic information and penetration control method in the welding process are studied.Finally,the reliability of the above method is verified by actual GTAW welding test.The specific research contents of this paper are summarized as follows:(1)The GTAW welding process penetration state detection and control test system is established.The test system integrates welding robot,automatic welding power,wire feeder,current and voltage detection sensor,molten pool image sensor,infrared thermal image sensor,arc sound sensor and other modules.The test system established in this paper can not only detect and control the GTAW welding penetration state online,but also establish a database to store the test data in real time so that the welding state of the test system can be monitored remotely through Web client.(2)Viewing of the molten pool image information in GTAW welding process,this paper proposes a variety of molten pool image processing and feature extraction methods,which are the traditional image processing methods that can quickly detect the length,width and aspect ratio of the molten pool respectively;Snake model method with slower detection speed but higher accuracy and robustness than traditional methods;Finally,the corresponding visual attention mechanism is proposed for the molten pool image to obtain the region of interest,and then the part-based model widely used in the field of target detection is introduced,and the parameters suitable for the welding molten pool image are set and improved.The trained part-based model achieves good results in feature extraction of GTAW molten pool image in terms of detection time and robustness against arc light interference.(3)For GTAW welding process arc sound information,we have studied the arc sound signal feature extraction method,after plenty of welding process test,we determine the arc sound information of time domain and frequency domain,RMS,standard deviation,energy and the average amplitude and short-time zero crossing ratio and zero than and arc sound and standard deviation of wavelet packet frequency band,and the frequency band energy,This 3-d sound arc characteristics for GTAW welding process on the back of the fusion width and penetration state’s most sensitive,so this paper in the arc sound signal in time and frequency domain signal analysis and the extraction of the above characteristics,to extract the characteristics of arc sound dimension information is overmuch.For the characteristic information redundancy problem,we have proposed the solution based on T-SNE manifold learning dimension reduction method in this paper,The 34-dimension arc sound characteristic information is reduced to 3-dimension.(4)GTAW welding process is a typical thermal process,and the temperature information of the welding process is an important information reflecting the depth and penetration of the weld.The existing research results of online detection of GTAW welding penetration at home and abroad rarely involve multi-information fusion combined with infrared thermal image information.In this paper,infrared thermal image temperature information is integrated into the multi-information detection method of GTAW welding penetration state,and the best collection angle of thermal image information in the GTAW process is tested and explored.After determining the best collection Angle of infrared thermal image information,Through theoretical analysis and welding process test,the highest temperature point and the surrounding point of the temperature in the heat affected zone of the molten pool are determined as the characteristic information points reflecting the different welding penetrations.(5)Status of the weld penetration tests are targeted design process test in the GTAW welding process,it does not involve levels millions or even tens of millions of huge amounts of data,so the research hot spot like deep learning and other multiple information fusion modeling methods are not adapted to the welding process on the back of the welding width and penetration state predicting dynamic modeling.Targeting relatively fewer but better test data in the welding process,this paper puts forward the improved particle swarm algorithm parameters optimization of support vector machine(SVM)method for a variety of sensors to extract the feature information of characteristics of the layer of information fusion or GTAW welding process dynamic model to predict the back weld fusion width and GTAW welding penetration state.Meanwhile,this paper makes a deeper exploration of GTAW multi-information fusion method,studies the application of Dempster/Shafer(DS)evidence theory method in the judgment of welding penetration state,and proposes DS evidence theory method of fuzzy rule optimization.A good solution is proposed to solve the decision layer fusion conflict in GTAW welding penetration state prediction.(6)The ultimate goal of GTAW welding penetration state prediction is to achieve a better effective control of the welding process to improve the forming quality of GTAW welding.Since it is difficult to establish an accurate mechanism model for GTAW process and there is obvious lag in the GTAW welding thermal process,model-free control does not need an accurate system mechanism model,while predictive control has good tracking performance and strong anti-interference ability and can directly deal with the characteristics of the process with lag.In this paper,control method relativing the innovative combination of predictive control and model-free control is put forward,we comprehensive their respective advantages and put forward the model-free adaptive predictive control method,at same time we apply it to real-time control of penetration in the GTAW process,simulation results and the final welding test show MFAPC(model-free adaptive predictive control)method which we proposed has obtained the good result for welding quality control.