Online Measurement And Quality Assessment Of Weld Surface Forming Based On Laser Vision

In modern industry,welding is an important processing and manufacturing technology which is widely used in the industrial field,the quality of welding has important influence on the reliability and safety of welding products,the post-weld quality inspection of welds is an important means to ensure weld quality.The line laser vision sensing technology is widely used in weld seam quality inspection because of its simple structure,strong anti-interference and high precision.Aiming at the effect of image noise on measurement accuracy in weld forming quality inspection process and the limitations of existing weld feature detection methods,the feature extraction of weld seam based on scanning laser vision sensing and the surface quality inspection of weld seam is studied.First,a surface quality inspection system of weld seam based on scanning laser vision sensor is designed,selecting reasonable hardware parameters and calibrate the vision sensor according to demands of system.By analyzing the distribution characteristics of gray-scale at the cross-section of the laser fringe,a weighted gray centroid method was proposed to extract the center stripe of weld seam.To reduce the influence of noise in weld stripe image on the accuracy of weld formation measurement,by analyzing the distribution characteristics of noise in weld image,a wavelet denoising method based on Soft-hard Threshold Compromise method is proposed.The wavelet multi-scale decomposition is applied to the center strip of weld seam to obtain the detail components of each layer at the center stripe of weld seam,by setting threshold for each layer detail component,detail component of each layer is corresponding retained or shrank according to the threshold,using the wavelet inverse transform to reconstruct the center stripe of weld seam with denoising.By analyzing the characteristics of stripe images with different types of welds,the corresponding feature detection algorithm is designed.For butt welds,by analyzing the deficiencies of the general slope method and intercept method,an improved algorithm based on the slope intercept method is proposed,multi-area detection of weld profile data is carried out to extract weld feature points.For fillet welds,when pits or splatter occurred on the surface of weld seam,the accuracy and robustness of the weld feature points extracted by Shi-Tomasi corner detection method is deficiencies,therefore,a weld feature point detection method based on segmented intervals is proposed,the proposed method has overcome the deficiencies of traditional weld feature detection methods by setting multiple thresholds to determine the position of the weld feature points.In order to realize the automatic detection of weld surface defects,the laser stripe images of different weld defect were analyzed.According to the geometry and spatial distribution characteristics of defects in the stripe profile,the feature extraction method of defects is studied and designed.According to the characteristic values of different weld defects,a weld defect classification model based on BP neural network was designed.The effectiveness and feasibility of the classification model were verified by experiments,experimental results showed that the established identification model can effectively identify surface defects such as pits,pores and undercuts in the weld seam.