Study On Algebraic Feature Extraction Of Arc Welding Pool Image Based On2DPCA Method

This paper first use the2DPCA method for feature extraction in the welding pool image, Then a classification system is designed based on the features of welding pool images which corresponding to the different forming quality. It can easily classify the welding forming quality through the welding pool image. In the2DPCA processing, the global data of images join in the feature extraction; the correlation between the images can effectively be eliminated, and features show big differences among types.In welding, misalignment, leakage, bad forming, slag inclusions are the common problem. Based on these, six types weld processing experience are designed, which are good forming, bad forming, misalignment left, misalignment right, leakage, and slag inclusions. And the corresponding welding images are captured by the passive optical monitoring. Discussing the feasibility and applicability of weld pool images in the three classifications by using the feature exaction algorithms called PCA,2DPCA,2D’PCA,(2D)2PCA which are based on principal component analysis. The experiments showed that the methods of2DPCA and2D’PCA win in the recognition rate and computational complexity for the feature extraction and classification to the welding pool images.A welding pool image classification software system is developed based on the MAG welding forming using the2DPCA method. The system includes two functions and three modules. The two functions are2DPCA and2D’PCA; three modules are welding pool image preprocessing module, training module of sample, discriminate module of welding pool image. All the preprocessing will be done in the preprocessing module; Training module will accomplish the calculation and storage of optimal projection axis and sample features; discriminate module is to discriminate which part the unknown image equals to. Experiments show that, for the series images of the bad forming in welding process5L/min shielding gas flow rate, the results of2DPCA are87%discriminated to bad forming,13%to leakage, while that of2D’PCA are75%to bad forming,25%to leakage; for series images of the good forming in welding process20L/min shielding gas flow rate, all the images on both methods discriminate to good forming; for series images of the good forming in welding process10L/min shielding gas flow rate,99.4%are defined good forming in the both method; among series image of the5mm misalignment right, the correct recognition rate respectively in two methods respectively are97.6%and98%.In summary, the classification system can effectively distinguish the forming quality by the welding pool image which design by the welding pool image quality feature based on2DPCA method. The method is easy to achieve and has a high recognition rate.