Intelligent Assmbly Detection System Based On Petri Network And Deep Learning

With the rapid development of economy,China's manufacturing industry is also developing rapidly,the number and variety of products increased,resulting in people's product quality requirements are increasing day by day.Because of the shortcomings of manual detection,such as fatigue,low efficiency,limited spatial and time resolution of the human eye,easy to produce false detection and leakage,so the use of automatic visual based on detection technology is more and more extensive.In the study of the existing machine vision based on assembly inspection system,it is rarely taken from the point of view of reachability,boundedness and deadlock-freeness of the whole system,the closed-loop inspection process between the assembly parts from the grasping action to the assembly results is considered,as well as adapting to the complex and changing part types and assembly environment.Because deep learning can use huge data compared with traditional visual detection,extract the rich features of the same target to complete model training,the robustness and generalization ability of the algorithm is stronger.Therefore,in view of the shortcomings of the existing assembly inspection system,and combined with the needs of manual assembly tasks,this paper establishes a deep learning based on motion recognition and assembly results detection system through the Petri network,respectively,the assembly personnel's grab action and assembly results are detected in real time,and the results of each test are interactively transmitted and analyzed between the two systems to achieve its closed-loop detection,and finally to improve the system detection efficiency and accuracy of the SSD algorithm,and to further improve the speed of model detection.The improvement of the SSD model is optimized and accelerated by Open VINO to better assist assembly personnel in quickly identifying leaks,misassembles and other problems in complex assembly environments.The details are as follows:(1)Through the analysis of the existing assembly inspection system,and combined with the needs of manual assembly tasks,the motion recognition system based on deep learning and the assembly results detection system are established,the call rules and behavior logic control of the software modules between the systems are analyzed,and the hardware such as light source,camera and lens is selected to realize the overall layout of the intelligent assembly inspection system.(2)In order to verify the accessibility,boundedness and deadlock-free nature of the system built,this paper takes a propeller assembly process as the research object,analyzes the entire assembly process of the propeller and the mutual constraint relationship between the parts,uses petri network to model the software module call rules and behavior logic control of the propeller assembly detection system,and then analyzes the performance of the built system model by using the correlation matrix and state equation analysis.Thereby,the feasibility of intelligent assembly inspection system based on Petri network and deep learning is verified.(3)In order to improve the system generalization ability and the accuracy of small target recognition,the characteristics of common target detection algorithm are analyzed and evaluated,the speed and accuracy of detection are comprehensive,and the SSD algorithm is used as the basic algorithm of system visual detection.Then the network structure and working principle of SSD algorithm are analyzed,and the algorithm is improved according to its characteristics of insufficient detection ability of small target and not strong enough algorithm,and the residual module and Idea structure are introduced into the model's underlying structure for multi-channel branch fusion,and the layer-jumping connections are made between the various levels,the different layer features are fused,and then the high-resolution and high-level feature high semantic information of low-level features are used through upper sampling to predict between different levels.Finally,the validity of improving SSD algorithm is verified by experiment.(4)Based on the Petri network model of the built system,the Petri network structure is modularized according to different functions,and the software architecture of the intelligent assembly detection system is established according to the mutual cooperation between the modules.Then,in order to improve the inference speed of the model deployed on the CPU,Open VINO optimized and accelerated the improved SSD model,and finally implemented the programming of the entire software system with the help of Lab VIEW,thus systematically completing the research work on the intelligent assembly detection system.