Research On Automatic Detection Of Special-Shaped Castings And Intelligent Recognition Of Defect Map

Based on the research background of in-situ automatic ultrasonic phased array inspection of special-shaped welding load-bearing workpiece,due to the limited in-situ detection position in the detection process,the shape of special-shaped welding load-bearing workpiece is relatively complex,and there are extreme multi-reflection structural features.It is necessary to identify and locate the features to be inspected,plan the optimal detection path among a plurality of features to be inspected,and detect the multi-scale targets of welding defect maps in a complex environment.This study mainly focuses on the porous location identification and detection path planning of shape of special-shaped welding load-bearing workpiece,as well as intelligent identification of defect maps.Therefore,a multi-feature contour recognition and location algorithm based on point cloud is proposed,which can simultaneously extract and locate the hole features and cylinder features of special-shaped welding load-bearing workpiece.Furthermore,based on the above-mentioned features obtained through visual recognition and localization technology,a detection path planning method is proposed to search for the shortest detection path of the robot.It is also proposed that the Faster RCNN combined with the Res Net 50 can be applied to welding defect maps detection.Based on the problems of coexistence of multiple small target defects and multiscale target detection,a method combining deformable network and feature pyramid network with Res Net 50 is proposed to improve the detection performance of the algorithm for multiscale targets,especially small targets.Based on the efficiency and accuracy of candidate frame selection,K-means clustering algorithm and ROI Align algorithm are proposed,which can be used to customize anchors and accurately locate candidate frames suitable for welding defect data sets.In this study,firstly,the "EyeinHand" hand-eye automatic detection system is established;secondly,the robot vision system collects and processes the point cloud information of the area to be inspected,and then extracts the geometric center and axis direction of the hole to be inspected;thirdly,the best detection paths of multiple target detection points are obtained by traversing all detection paths through multi-branch tree;finally,the self-made ultrasonic atlas data set of weld defects and the improved algorithm proposed in this study are used to verify the target detection experiment.The experimental results show that the positioning accuracy of the feature to be measured by the vision system is 0.107 mm,the aperture extraction accuracy is 0.002 mm,the cylindrical surface fitting accuracy is 0.04 mm,and the error between the two axes of the angle calculation accuracy is within 0.4°.When the number of features to be inspected is different,the average moving distance of the end effector of the inspection path can be saved by 10.7% after the path optimization.The overall accuracy of type recognition is 93.46%,and the accuracy of small target defects such as " stoma " and "crack" are 91.6% and 87.7%,which is 4.54% higher than that of the original Faster RCNN algorithm.The ablation experiments and comparison experiments with other mainstream target detection algorithms prove that the improved method proposed in this study improves the detection performance.The feasibility of in-situ automatic ultrasonic phased array inspection of special-shaped welding load-bearing workpiece by visual positioning is verified through the verification of the actual industrial inspection site,which basically meets the requirements of weld defects inspection.This study content can provide a reference for the in-situ automatic ultrasonic phased array inspection method of special-shaped welding load-bearing workpiece.