Research On Predicting And Intelligent Control For Weld Formation During VPPAW Process Using Multi-information Fusion

Among the high energy beam welding technology which is known as “one of the most prospective advanced welding technology in 21 st century”,variable polarity plasma arc welding(VPPAW)has many adavantages of high energy density,strong penetrability and free-porosity which is widely applied in high-end manufacturing area such as aerospace,nuclear power etc.Due to unqiue physical attributes of plasma arc,it brings a series of problems in VPPAW process such as the stability of weld pool and keyhole,the sensitivy of welding technical parameters,and norrow process magarin.Therefore,how to realize the stable opearation of welding process,widen the magarin of process parameters,represent the keyhole behavior and weld pool information accurately and obtain fine control stability for weld formation has been one of the most challenging issues in welding field.At the background of welding intelligent manufacturing for space navigation field,this paper attempted to develop a new approach for intelligent processing of multi-source based on electrical signal,acustic signal,and visual signal,and online monitoring and real-time stability control for welding process.This paper deeply studied the weld formation and penetration control for 2219 aluminum alloy VPPAW process.An experimental system for VPPAW has been set up,which can automatically control the welding process.A multiple-signal acquisition system has been built as well in order to acquire and save the information from welding current,arc sound and keyhole pool images.In order to realize the monitoring for the dynamic variation of keyhole pool in VPPAW process,a dual-light-route visual sensing subsysyem was designed to catch keyhole pool images from top and back directions simultaneous.Through the analysis of the shape changing in weld pool-keyhole,a new image recognition method based on part-based model was developed to extract the visual features.Then the process parameters were adjusted to change the arc penetrating capacity,in order to observe the influence tendency of weld quality on pool visual features.Based on this analysis,this paper further studied the relationship between the keyhole dynamic variation and weld quality,and analyzed the change mechanism combing with the established “thermal-force” model.However,it is sometimes beyond possibility to acquire clear and stable visual images,which is affected by strong plasma arc,flame or splash.Considering the arc sound contains abundant information about welding dynamic process,this paper introduced the weld arc sound as another important sensing method.On the basis of understanding the unique “double-sound-sources” in plasma arc welding,it analyzed the welding dynamic process and penentration status based on modern signal processing method.In addition,this paper focused on the method and principle of the pulse signal feature extraction,including the aspects of time domain features,frequency domain,and mel-frequency cepstral coeffients.Fianally,the relationship between the arc sound features and back-side keyhole behavior and weld quality has been analyzed to provide the penetration features for monitoring penentration status and formation control in VPPAW process.On the basis of the Multi-source information fusion for keyhole pool visual images and plasma arc sound,this paper proposed a novel manifold learning called t-SNE,which can achieve the high-dimensional feature reduction and data visualization.Then taking the deep belief network(DBN)as a classification algorithm,a penetration classification model based on low-dimensional space after feature reduction has been constructed to qualitative evaluate the actual penetration status.At last,the paper introduced a penetration prediction model based on adaptive neuro-fuzzy inference system(ANFIS)to predict the back-side weld width at the current instant,in order to provide an accurate feature feedback for subsequent closed-loop penentration control.After the weld process information acquiring,fusion processing and knowledge modeling,this paper designed a model-free adaptive control(MFAC)to overcome the difficulity of building an accurate mathematic model of welding process.The MFAC takes the back-side weld width as the controlled variable to acquire uniform weld formation.To compare the controlling performance,this paper designed a simple-variable PID controller and a multi-variable intelligent controller to carry out a series of simulation and weld control experiments to prove validity of two kinds of controllers.controllers.The control results of single-variable PID controller with weld current as control variable showed that though it could prevent the welding process from burn though effectively,but could not get ideal uniform weld formation.The multi-variable intelligent controller could achieve ideal uniform and stable weld formation.Even under much heavy welding conditions of varied heat sink and varied heat input with controlling variables: weld current and plasma gas flow.The perfect and stable weld formation is achieved.