Research On Product Change Model Construction And Path Planning Based On Improved Dendritic Neural Network And DQN Algorithm

In recent years,with the complex external environment and changing customer demands,enterprises have increasingly demanded changes to their existing products.However,it is often difficult for enterprises to accurately grasp the relationship between product components and the propagation path of changes,leading to the need for engineering experiments on a large number of feasible change plans to find the optimal product change plan,resulting in a waste of manpower and material resources.Therefore,in order to save costs and improve efficiency in product changes,it is crucial to build product change models and plan the optimal path for product changes.However,existing studies on product change models are often based on expert knowledge and experience,ignoring the hidden knowledge in historical changes.In addition,most change path planning methods only consider the connection between product components and do not identify the impact of component changes on the overall product.In view of these issues,this thesis studies the problem of constructing a product structure network model and planning the change path,which mainly includes the following two aspects:(1)Product structure network model construction by fusing dendritic neural network and false connection recognition: Given the problems of subjective judgment in determining connection relationships and lack of effective utilization of historical change data in existing product change model construction methods,a product structure network model construction method was proposed that fuses dendritic neural networks and false connection recognition.Firstly,an improved dendritic neural network was proposed based on the characteristics of the design structure matrix and product structure network model to extract the implicit connection relationships and strengths between product components in historical change data.Secondly,by integrating depthfirst search and expert experience,the problem of false connections caused by the network’s inability to recognize change propagation paths was solved,and a more accurate product structure network model was constructed.Finally,a 2D bicycle model was used for experimental analysis,which showed that the product structure network model established by this method could accurately reflect the association relationships between product components.(2)Product change path planning based on change impact indicators and DQN algorithm: Based on the product structure network model proposed in(1),a product change path planning method based on change impact index and DQN algorithm is proposed to further improve the efficiency of finding the cost-optimal product change plan.Firstly,a node change impact index is proposed to characterize the impact of component changes on the overall product,and it is used together with node change cost and product structure network to construct a product change simulation environment.Then,based on the characteristics of product change propagation,the value function network,environmental state,agent action,and reward function in the DQN algorithm are redesigned.Finally,based on the product structure network model established in(1),different nodes were selected as initial change requests for corresponding experiments to verify the effectiveness of the proposed product change path planning algorithm.The dissertation has 31 figures,17 tables,and 81 references.