Feature Detection And Compensation Control Of Porosity During AC-GTAW Of Aluminum Alloys Based On Arc Spectral Data

Aluminum alloys have promising application in automobile and aircraft industries due to their high specific strength and good corrosion resistance.However,porosity is easily produced during welding solidification due to the significant difference of hydrogen solubility in liquid and solid alloy,which is detrimental to welding quality.Because of existence inside the weld seam,porosity has to be detected by destructive test or nondestructive test,which is hard to meet the need of efficient production with high quality for modern manufacturing.As an associated signal in welding process,arc spectrum contains large amounts of information of different elements relating to the welding dynamic process.And understanding the internal relation among porosity,welding parameters and arc spectral data is the premise of finding the relationship between spectral features and porosity.Therefore,the paper studies the formation mechanism of porosity and develops on-line monitoring technology of welding quality based on arc spectral information,aiming at exploring a new approach for real time predicting inside porosity and assuring the quality reliability of welding products.In order to detect porosity defection in pulsed GTAW of aluminum alloys,spectral information collection system of automatic welding process is built.Moreover,spectral sensing system based on linear CCD is designed for real time collecting and processing data.The formation mechanism is investigated by bubble nucleation and bubble growth.Models of spontaneous nucleation and heterogeneous nucleation on the surface of refractory materials are established respectively,leading to the conclusion that the energy needed for spontaneous nucleation is less than heterogeneous nucleation.However,the critical nucleation radiuses in both models are the same and vary inversely as the square of the welding current.Furthermore,the growth trend is analyzed based on the model of spontaneous nucleation.The result shows a single porosity presents turbinate.An improved K-medoids algorithm is proposed to accurately extract the emission lines of spectrum for spectral signal analysis.It uses center points to get the corresponding linear line intensity,which provides accurate data source for the calculation of electron temperature( T_e )and intensity ratio.And in this paper,an new similarity measure function-spectral distance is defined to replace the traditional Euclidean distance according to the variation characteristics of spectral data.During the following manifold learning,the spectral distance is used to calculate the distance matrix.By boltzman method,electron temperature T_e is calculated based on clustered hydrogen line.And compared with infrared image,the calculation result of T_e satisfies the temperature distribution of the arc.The continuous electron temperature T_e curve of hydrogen lines is affected by the power and shows the pulse period fluctuation.The effect of pulse signal can be removed effectively by wavelet packet decomposition and then we get features characterizing defects with obvious appearance.But for internal gas pores,it still needs to analyze the hydrogen/argon intensity ratio signal.The ratio signal has non-stationary and non-linear characteristics,so empirical mode decomposition is deployed to isolate high-frequency pulse and noise signal.The reconstructed signals are more detailed and computation time is shorter compared with wavelet packet analysis.After analyzing the local atomic spectrum lines,the whole spectrum data is studied by manifold learning.It is found that the nonlinear dimensionality reduction method can better reflect the data structure of the spectrum than the linear method.The study result shows the inner relationship between components after reduction and defeats of gas pore.To solve the problem of locally linear embedding algorithm that it is unsupervised and can’t deal with the point outside samples,an improved algorithm is proposed based on maximum edge criterion,which makes the classification effect under 3D visualization obviously improved.For getting highest pore classification accuracy,six parameters including average,range and kurtosis are extracted from the hydrogen spectrum.And genetic algorithm is used to obtain the best penalty factor c and the kernel parameter g,and the classification accuracy of the pore is increased to94%.Based on extracted spectrum features,the effects of different welding conditions on the pores are further studied.From groove angle,butt gap and thickness of the welding root these three conditions,we made the real-time acquisition of spectrum data to predict the presence of gas pore defects,and the X-ray nondestructive testing is used to verify the correctness of the prediction.Then the orthogonal test was carried out by the above three factors,and the optimal combination with the lowest porosity was found.Finally,on the basis of fuzzy self-adaptive PID control,control strategy of restraining porosity is obtained by adjusting welding current to keep the spectral intensity ratio at the ideal value and is verified to be valid by experiments.