Research On Pattern Recognition Algorithms And Real-time Detecting System For Welding Defects System Base On The Infrared Vision

With the developing of national defense and manufacturing industries,the quality assurance of weldments are keeping increasing.In view of the fact that traditional nondestructive testings methods（X-ray,ultrasonic,magnetic powder,etc.）have many disadvantages such as heavy work intensity,high costs,no real-time,in this paper the GMAW arc welding and laser welding are conducted,the research of welding visual information acquisition,image preprocessing technology and recognition algorithms for welding defects are studied.The pattern recognition system for welding defects is completed preliminary,and the accurate identification for 4 common welding defects（shape defects,overheating,spatters and non penetration）is realized.The main contents of this paper include the following points:To reduce the interfere of arc,the infrared CCD（wavelength:1-5μm,frames:10-500 Hz）was adopted for the images acquisition of molten pool.According to the noise characteristics in images,an improved filter algorithm and an improved bilinear algorithm were used to preprocess the images,and post-images with clear features were obtained.The temperature calibration of infrared image was done by measuring the real temperature of the molten pool via tungsten-rhenium thermocouple,the relationship between real temperature T and the gray value of pixel x of image was fitted as the formula:T=1718-1788.4/(1+e^{（x-110.6）/28.1}),where the Adjusted R Square is 0.995.The width of molten pool,welding center-line and area ratio of nugget zone to molten pool were obtained via threshold segmentation processing of image.The characteristics of defects in GMAW welding（shape defects,overheating,spatters）and non penetration defect in laser welding were analyzed and summarized for pattern recognition,which provided the ideas for the following algorithms design.For GMAW arc welding,a)A recognition algorithm was designed,which detecting shape defects by calculating discontinuous zones of molten pool,variances and derivative of width,residual error of fitting for weld center-line.b)A recognition algorithm was designed,which detecting overheating defects by analyzing maximum width,maximum length,the area of nugget zone.c)An algorithm for detecting isolated regions based on position iteration was designed,which can find out the numbers and area of spatters from molten pool images.For semiconductor laser welding,the relationship between the temperature dropped in 2.5 s after the end of welding and the penetration of welding seam was established.An algorithm was designed,which can detect non-penetration defects by analyzing the area ratio of nugget zone and molten pool.A welding defects recognition system software was developed,the interface is friendly and can be used to monitor the welding process in real time and identify the welding defects,providing a new approach for nondestructive testing of welding quality in manufacturing production.