Study On Identification Of Weld Deviation And Penetration State In K-TIG Welding

Medium-thick plates can be single pass welded with double-side formed without opening grooves by Keyhole Tungsten Inert Gas(K-TIG)welding,which has high welding efficiency and significant advantages in the field of medium-thick plate welding.However,real-time monitoring of the K-TIG welding process still needs to be studied in depth.In this investigation,identification of the weld deviation and penetration state in K-TIG welding were studied,aiming to lay the foundation for real-time seam tracking and control of weld penetration state,which has important research significance and application value.In this investigation,the weld deviation identification system based on visual sensing and the weld penetration state identification system based on acoustic sensing were constructed,respectively.On this basis,the appropriate system parameters were obtained through experiments,and the camera and the microphone were separately calibrated.At the same time,a set of K-TIG welding arc acoustic signal acquisition and weld penetration state identification software based on LabVIEW and MATLAB hybrid programming was developed.In order to identify the weld deviation,a CCD camera with high dynamic range(HDR)characteristics was used to capture high-resolution welding images.For the characteristics of strong arc interference and narrow-gap seam during K-TIG welding,an improved regiongrowing method was proposed to extract the arc center,and the edge fitting algorithm based on sampling decision was proposed to extract the centerline of the weld,and then the distance between them was calculated,which was defined as the weld deviation.It has been confirmed by experiments that when the welding trajectory was along centerline of the weld or deviated from the centerline of the weld by a certain angle,the accuracy of the weld deviation identification algorithm was within ±0.30 mm,which meets the requirements of industrial production.In order to identify the penetration state of the weld,experiments were designed to obtain the arc acoustic signal library under different welding current,welding speed and the distance from the tip of the tungsten needle to workpiece(CTWD)during K-TIG welding.And the relationship between the arc acoustic signal and the weld penetration state was pointed out by analyzing the backside keyhole image.The arc acoustic signal was analyzed in the time domain,frequency domain and timefrequency domain,and the ability of 36 different acoustic characteristics to characterize the weld penetration state was studied.The analysis of a large number of data shows that some acoustic characteristics such as the relative energy of the wavelet packet in the frequency band of 2.5~5.0kHz,can better characterize the penetration state in K-TIG welding.The feature selection method that fuses filter and wrapper method was then proposed,and finally the 12-dimensional optimal feature subset which can better characterize the penetration state was obtained.The Support Vector Machine(SVM)penetration state identification model was established by taking the optimal feature subset extracted from the arc acoustic signal in K-TIG welding as the input and the three weld penetration states as the output,and the identification accuracy of the model can reach 92.21% after parameter optimization.The model was compared with the Extreme Learning Machine(ELM)and Back Propagation Neural Network(BPNN),and the results showed that the SVM model had better identification performance on the dataset.It has been confirmed by experiments that the weld penetration state in K-TIG welding can be identified in real time based on the developed software and the SVM model.