On-line Detection And Evaluation Of Arc Welding Based On Spectral Analysis

In industrial production,arc welding structure is an important connection part of all kinds of mechanical structures,and weld quality is of great significance to production safety.At present,there are many methods for weld quality inspection,but most of them are carried out after welding,and there are low efficiency and high cost,so it is of great significance to study the on-line inspection and evaluation method of welding process to improve the efficiency of welding quality inspection and welding on-line guidance.There are many signal sources in the process of arc welding,among which the arc spectrum signal is accompanied by the whole process of arc welding,which contains many important information of the whole welding dynamic process and is closely related to the weld quality.Therefore,this paper takes the real-time spectral signal in the arc welding process as the research object,studies the plasma spectral diagnosis and spectral data processing methods,analyzes and excavates the relationship between the arc spectral information,welding process parameters and weld quality,and effectively realizes the defect detection in the welding process.Specific studies are as follows:(1)Based on the experiment,the arc welding process under different process parameters and its influence on the formation of porosity defects were studied.Based on the welding test platform configured by arc welding robot and signal acquisition system,changing the controllable process parameters such as welding current,protected gas flow rate,welding speed,etc.,set up a control test to analyze the influence of different welding process parameters on spectral strength,weld morphology,and the formation of porosity defects.The experimental results show that the porosity decreases with the increase of welding current and decreases with the decrease of welding speed.(2)A k-medoids clustering algorithm combined with synthetic similarity measure is proposed to obtain the spectral intensity of spectral characteristic lines accurately.For solving the problem that the peak intensity of spectral characteristic lines is not unique due to wave line broadening and instrument zero drift,this paper proposes to describe the classification process of spectral data k-medoids clustering algorithm based on minimum spanning tree,so as to obtain the spectral intensity of characteristic lines accurately and lay the foundation for the accurate calculation of electron temperature of characteristic lines.To optimize the classification processof spectral data by clustering algorithm,a comprehensive similarity measure function is constructed to calculate the distance of spectral data points to further study the variation characteristics of spectral data.The results show that the cluster synthesis evaluation value of the integrated similarity measure function in the bands of 661.57 nm~762.24 nm is 0.8787,and the cluster synthesis evaluation value in the bands of 762.64 nm~941.57 nm is 0.868,which is higher than the cluster synthesis evaluation value of the three similar measure functions,spectral angle matching,spectral information divergence and spectral correlation.(3)The feature extraction method based on wavelet packet transform for the characteristic spectral line electronic temperature curve is presented,and the effective characterization of weld surface defects is realized.According to the selection criteria of plasma spectral lines,the characteristic spectral lines of Argon atoms and their spectral intensity are obtained by clustering algorithm,and the electronic temperature curves are calculated by Boltzmann method.In order to eliminate the influence of peak current and base current on the change of electronic temperature curve,wavelet packet transform is proposed to extract the characteristics of electronic temperature curve.The amplitude threshold matrix of each frequency band of the reconstructed signal can better describe the surface defects of the weld.(4)A local linear embedding algorithm is proposed to realize dimensionality reduction and feature extraction of nonlinear spectral data.Three linear data dimensionality reduction methods,PCA,MDS and LDA are studied.The visualization results of data dimensionality reduction processing show that the dimensionality reduction effect is not ideal.Aiming at the deficiency of linear dimensionality reduction method and the nonlinear data features of massive redundancy,we propose to use local linear embedding algorithm to extract spectral information features.The spectral intensity and intensity ratio of spectral line Ar I 794.81 nm?Ar I 840.83 nm of spectral features can characterize porosity defects.To further improve the prediction rate of welding defects,the accuracy of defect prediction is increased to 95% by using decision tree algorithm.