Research Of Welding Line Detection Algorithm Based On Structured Light

Ultrasonic non-destructive testing(Ultrasonic NDT)is one of the important techniques to ensure the quality of metal weldments.In practice,NDT of weld seam in pipes or intersecting lines is performed manually by the professional technician.In the harsh working environment and high-intensity working conditions,the data captured by manual NDT not only has large deviation but also poor consistency and low efficiency of manual work.Therefore,automatic NDT technology has emerged with the goal of automatically detecting the weld line and guiding the ultrasonic probe along the weld line for scanning.In this paper,a parallel structured light(PSL)platform for wall-climbing robots is designed,and a weld line detection scheme based on structured light is proposed to guide the robot to identify and track the weld line.The PSL platform is equipped with two cross-projected structured light generators,each of them projects five parallel structured light stripes at the weld boundary.A camera that captures the structured light image is also installed on the PSL platform,and then an image processing algorithm is proposed to identify the angle and distance between the weld line and the robot.Image processing in this paper includes image preprocessing,stripes feature extraction and stripes feature analysis.(1)Image preprocessing,using the SSD-MobileNet model to locate and identify the region of interest(ROI),extracting two ROIs with an area of approximately 10%of the original image Image processing according to ROI not only avoids the influence of light pollution,noise and other disturbances in non-target areas,but also improves the processing speed of the algorithm and ensures the quality of the pre-processing.(2)In stripes feature extraction,the gray centroid method is used to extract the center line of the stripes,with the extraction accuracy is sub-pixel level.Then,through the improved synchronization region growth algorithm and interpolation algorithm,all center points are classified into different stripes and the local disconnection has been repaired.(3)Stripe feature analysis,this paper proposes a stable and fast weld boundary point extraction method based on the stripe geometric feature.Then two different weld line recognition algorithms are proposed,which are iterative-based least square weld line fitting algorithm and information fusion based global weld line search algorithm.The former utilizes the distribution variance of the boundary points to eliminate the interference points and optimize the weld line fitting effect,while the second method constructs a weld line search model based on global boundary information,establishes a cost function,and then finds the optimal solution by traversing all possibilities.Therefore,the second method considers the information correlation between multiple weld boundaries,and the experiment proves that the second scheme is more accurate and stable.Finally,a verification experiment was carried out,and the ROI prediction accuracy by SSD-MobileNet model reached over 90%.The results of 18 sets of identification and tracking weld lines in pipeline welds and Y-shaped intersecting line welds show that,the distance tracking accuracy is 0.6mm,and the angle tracking accuracy is ±1°.The experimental results meet the design requirements of this paper and the actual Ultrasonic NDT requirements.