Research On Flexible Process Planning Method Via Knowledge Graph And Reinforcement Learning

In the information age,product demand tends to be personalized,diversified and dynamic.The market of small-batch customization products develops rapidly,but it has some problems in the process of machining,such as difficult process resource organization,low process planning efficiency and poor dynamic adaptability.Therefore,an intelligent process planning method based on knowledge graph and reinforcement learning is proposed.Under this method,three key technologies involved in flexible process planning,including process knowledge representation and management,candidate set generation of manufacturing resources,and process route optimization,will be studied in order to meet the requirements of dynamic process planning for smallbatch customized products.In terms of process knowledge representation and management,firstly,concept relations in the process are sorted out according to the seven-step ontology construction method,and the process domain ontology is constructed using Protégé tool.Then,process knowledge entities are extracted based on BERT-Bi LSTM-CRF model,and relationship establishment and entity alignment are further completed.Thirdly,ontology is used as the pattern layer of knowledge graph to guide entities and relations to fill the data layer.Finally,the knowledge is stored and visualized based on graph database Neo4 j,and an intelligent construction method of process knowledge graph is established.In the generation of manufacturing resource candidate sets,first of all,the low dimensional vector is trained for the entities and relations in the process knowledge graph via knowledge representation learning technique.Then,a processing chain auxiliary generation method is proposed,which calculates the vector similarity of multiple attributes,finds similar features for the target features to achieve the reuse of the processing chain,and provides auxiliary decision-making knowledge for the features of the processing chain without reuse.Finally,a qualitative and quantitative double-layer filtering method is used to match manufacturing resources for each processing element in the processing chain,and the final candidate resource set is obtained,which is used as the data input for process optimization.In the process route optimization,a multi-objective optimization model is established,which took the total process cost,total time and total carbon emission as the optimization objective and the priority relationship between processing elements as the constraint.Then,the optimization problem is transformed into markov decision process,the state space and action space are defined,and the reward function is set by using hypervolume as multi-objective evaluation index.Finally,a deep reinforcement learning method based on actor-Critic structure is designed to solve the optimization model.In the case study,the feasibility and superiority of the flexible process planning method based on knowledge graph and reinforcement learning are proved by comparing the solution effect between single and multiple objectives and the optimization ability between the method and heuristic algorithm.Furthermore,it is verified that this method can cope with the decision-making requirements of dynamic changing process conditions.