Design Of Virtual Assembly Process For Mechanical Assembly Based On Agm-vmm Neural Network Algorithm

With the rapid development of artificial intelligence technology and digital technology,the application of intelligent algorithms in virtual assembly has made rapid progress.Using intelligent algorithms such as genetic algorithm,ant colony algorithm and neural network algorithm to design virtual assembly has long become an important research direction in the field of virtual manufacturing.Compared with other intelligent algorithms,neural network algorithm has gradually become an important research direction of assembly planning research due to its good ability of information acquisition and knowledge learning.However,because the neural network algorithm is easy to fall into local optimal solution and training efficiency is slow,the application effect in virtual assembly is not satisfactory.In addition,in the aspect of assembly sequence planning,the existing methods seldom analyze the attribute information of parts themselves and assembly process information.In the aspect of assembly path planning,the current research mainly focuses on two-dimensional plane path planning,but there is still a large gap in the research of three-dimensional space path planning.In view of the above mentioned problems,this paper carries out the following research work:In order to improve the training efficiency of neural network algorithms and reduce the training cost,In this paper,Adaptive Gradient Method(AGM)and Vector Memory Method(VMM)are proposed to improve the neural network algorithm based on the analysis of the loss function information and parameter updating trend information in the process of neural network algorithm training.Among them,AGM increases training incentive by introducing parameter correction term,and corrects the updating of weight parameters.VMM reduces the time complexity of the algorithm by establishing a gradient experience reference and loss function detection mechanism.The validity of the two optimization strategies is verified by experiments on the XOR problem,pattern classification problem and handwritten digit recognition problem.In order to realize intelligent assembly sequence planning under multi-constraint and multi-objective conditions,this paper studies the assembly sequence planning method for assembly component attribute information and assembly process information.A feasible assembly sequence deduction model was built based on collision interference and sequence inversion.By adding disassembly level and priority direction conditions of parts and assemblies,the artificial experience knowledge was transformed into constraint conditions for solving assembly sequences,which reduced the sequence solving space and improved the computational efficiency of the algorithm.The assembly sequence knowledge learning model was built based on the neural network.By learning the parts attribute information known to the optimal assembly sequence,the hidden association knowledge was extracted.Based on the above two research models,a composite evaluation function of assembly sequences based on assembly attribute information and assembly process information was established to realize the comprehensive evaluation of assembly cost of assembly sequences.The effectiveness and rationality of the proposed method for assembly sequence planning are proved by the simulation experiment of Parallel-jaw vice and aeroengine.In order to realize real-time and efficient assembly path planning of parts in unknown space environment,a three-dimensional assembly path planning method based on environmental information and motion information about parts is studied in this paper.The three-dimensional spatial environmental information model of parts was constructed,and the environmental information was discretized into a finite state space to reduce the information dimension.The local obstacle avoidance model based on neural network was constructed to extract and analyze the obstacle information in the environmental information.The cost analysis of path planning based on kinematics was studied.The heuristic rules of path planning were proposed by analyzing distance cost,steering cost and path coincidence incentive,and the comprehensive selection probability model and decision rules of motion decision were constructed to realize real-time and efficient processing of path planning information.By comparing the proposed method with A* and Navmesh algorithm for path planning,the feasibility of the proposed method is verified.Finally,the virtual assembly simulation platform is designed based on Unity3 D,Visual Studio,My SQL and other software.Based on the modular design idea and structural design criteria,the system structure is designed to improve the portability and applicability of the platform,and the human-computer interaction function,virtual assembly simulation and other functions are designed,and finally the system is released on the PC end.