Research On Visual Recognition And Location For Industrial Robot-Enabled Disassembly For Products In Remanufacturing

As industrial robots are introduced into various complex industrial environments as an advanced intelligent manufacturing equipment,higher requirements are placed on the intelligence,accuracy and stability of robots.As one of the core technologies of environment perception for robot,machine vision can improve the autonomy of industrial robots and enhance their flexibility.Therefore,it has received great attention in research and production field.Although some achievements have been made in the field of industrial robot-based remanufacturing disassembly,most of the methods are traditional image recognition algorithms based on artificial feature extraction.The selection of features still depends on professional knowledge,and the robustness is poor in complex environments,so it is difficult to achieve effective object detection.This paper focuses on the recognition and location of industrial robots in the disassembly and remanufacturing environment.In view of the unstable structure of the object to be disassembled and the complex disassembly environment,the multi-objective detection algorithm with deep learning as the core is deeply studied to provide technical support for intelligent disassembly of robots.The main research content of this paper is as follows:(1)Research on multi-objective recognition and localization algorithm for mechanical and electrical products disassembly on industrial small data sets.The state of the art object detection algorithms based on deep learning are analyzed,and some methods based on region proposal are compared with the end-to-end object detection methods.Aiming at the low detection accuracy of small targets in the disassembly process,the Single Shot Multibox Detector detection algorithm is selected to improve,and a light convolutional network structure combining the low layers' detail features and high layers' semantic features is proposed.According to the distribution of the data set,the aspect ratio of the anchor is enriched to improve the detection accuracy.In order to solve the problem that the industrial data set is small and easily lead to network over-fitting,it is proposed to improve the network convergence rate and robustness through data enhancement and fine-tuning.(2)Monocular stereo vision positioning for industrial robots on mechanical and electrical product disassembly.In order to realize the transformation of the disassembled parts from the two-dimensional image coordinates to the three-dimensional space coordinates,the coordinate system transformation,the camera imaging model,the monocular stereo vision principle,and the robot hand-eye calibration are studied.The monocular vision positioning system of industrial robot is constructed,and the camera parameters are obtained by checkerboard calibration method.Aiming at the inaccurate extraction of the corner points on the calibration plate,it is proposed to use the circular target replace the checkerboard target.Combining with hand-eye calibration technology,the parts of the disassembled object are positioned in space,thereby guiding the robot to complete the disassembly.(3)Design and implementation of industrial robot disassembly vision recognition and positioning system.In order to verify the effectiveness of the deep learning-based target recognition and localization algorithm proposed in this paper,based on KUKA industrial robot,Baumer camera and computer system,the workpiece recognition and positioning system in the process of industrial robot disassembly is designed and implemented.The system includes an image acquisition module,an image recognition and positioning module,a spatial positioning module,and a robot disassembly execution module.Experiment with the idler in the engine as a disassembly object to verify the feasibility of the system.