Recognition And Realtime Control Of Penetration Characteristics Of Al Alloy Pulsed GTAW Process Based On The Arc Audio Information

The real-time monitoring of welding quality has been more important and difficultyin automatic welding process. Traditional post-welding defects detection methods aredifficult to meet the requirements of modern welding manufacturing, which ischaracterized by high quality, high efficiency and low cost. The penetration state is thekey judgment of welding quality monitoring. Arc sound signal is one of the associatedsignals of welding process. It has been proved to be the effective and essentialinformation for weld quality control. The experienced welder can roughly judge thewelding dynamic process, especially penetration state only with listening. It means thatthe arc sound signal contains a large amount of welding dynamic process information.The information content of arc sound signal is as much as the visual sensor information.In order to use arc sound information and convert the imcomprehensible features to akind of “readable image”,it is necessary to understand its mechanism and the concretedistribution of arc sound feature. Therefore, this paper proposed a new way of on lineprediction and control for the weld quality based on modern signal processing methodand speech recognition theory.In this paper, the research was accomplished at the multi-information collection and welding automation control platform in the background of welding quality control foraluminum alloy pulse GTAW welding. The arc sound signal was set to be the researchobject. Firstly, the incentive mechanism of arc sound signal was observed by comparingwith the arc voltage and current signal frequency characteristic. The incentive of soundsource appeared periodic in70Hz, same as the frequency of welding power source. Itproved that the welding power frequency was the excitation source of arc sound signal.The change of arc energy was the key influence for the sound incentive. The inside ofexcitation source for arc sound signal had one primary peak and one sencondary peak.The primary peak was correspondent to the arc ignition, the secondary peak wascorrespondent to the arc extinction. Through the DCT analysis, the results showed thatthe primary peak during the arc ignition played an important role for the generation of arcsound signal, the sencondary peak during the extinction showed little impact for arcsound generation.In order to study the further characteristic of arc sound signal, a new way oftime-frequency-timefrequency domain feature extraction of penetration state wasproposed, including the auditory attention AC-ROI extraction preprocessing method andthe maximum modulus threshold denoising method, they could effectively remove thenoise and extract the most related information. Through the time-domain analysis,frequency-domain analysis and wavelet analysis, a23set vectors was set to bepenetration feature for arc sound signal, like mean sound>0.08，energy>100，standarddeviation>0.12，covariance>0.015，kurtosis factor>3，skewness factor>2，after DCTanalysis the frequency band information between5.5-9.5kHz could recognize differentpenetration state; the frequency band energy between3.75-5kHz,5-6.25kHz,6.25-7.5kHz and8.75-10kHz could recognize as the penetration state feature; thefrequency band tolerance between5-6.25kHz and8.75-10kHz could set to be thepenetration state feature; in addition, the penetration feature coefficient of arc sound signal channel were extracted likex (n)and ai. Under larger amounts of data analysis,all these charateristics have a good correspondence to the penetration state of weld seam.Considering the nonlinear relationship between arc sound signal and penetration state,a new forecasting model named BP_Adaboost NN was designed for predicting thepenetration state for welding dynamic process, the recognition rate was about94%. It hadthe advantage of high accuracy, high applicability. Also a new way of modeling wasintroduced and designed for forecasting the penetration state like WA-HMM forecastmodel. It combined the advantage of wavelet anyalysis and hidden markov model. Thenew model was a second-order differential MFCC with6state WA-HMM predictionmodel. The highest recognition rate is95.83%.Considering to the special characteristic of the sensitive response for arc soundchange to the arc length, a new way of modeling was proposed to set a prediction modelbetween the arc sound signal and arc length, also with the collapse of weld. It contained anew way of denoising method named wavelet packet-moving average filtering method. Itcould successfully remove the environmental noise and the pulse interference noise fromthe arc sound signal; a new way of piecewise-linear fitting method was proposed toachieve the piecewise arc length prediction in two range4-3mm and6-5-4mm. Theeffective arc length was set to range from (3-7)mm considering the special character ofpulsed GTAW. The prediction accuracy was0.580487mm much higher than the singlelinear model. It was also divided into different step model for refinement like1mm stepmodel and2mm step model. These two step model were designed for two situation likearc length slowly changing process under low current and arc length rapidly collapsingunder large current.Finally, a new software system was developed to analysis the pulsed GTAW arcsound signal and modeling. It contains: load module, data preprocessing module, feature extraction of arc sound signal module, auxiliary visual image processing module,penetration state indentification module and arc length prediction module. It provided aconvenience tools for the analysis the arc sound signal. A new way of controlling thepulsed GTAW dynamic process was designed and verify. It contained two parts: one wasthe arc length controlling part, the piecewise-PID controller could achieve the real-timecontrol of arc length and weld collapse during the welding process. The PW-BP controllercould achieve the real-time closed loop control of weld penetration welding experiment.The results showed that comparing to constant welding parameter experiment, the weldquality was improved obviously and forming quality was good.