Construction and Application of the Knowledge Graph Method in Maintenance of Robot in Automotive Manufacturing Industry

摘要

Based on the spare parts and structure data of industrial robots, the entity list of robot parts is established to form a query dictionary, and entity annotation is performed on the robot corpus by means of dictionary query, which reduces the cost of manual annotation and ensures the quality of annotation data. In the process of entity recognition training, Bert+Bilstm+CRF model structure is used to initially use 70% of the dictionary data for annotation, and the model is trained by iteratively increasing the annotation data in a continuous cycle, so that the model can extract all the entities in the robot corpus as much as possible. In addition, the material number/model information and PM maintenance content/strategy of the entity have been used as attributes of the entity. Meanwhile, the experience summarized by the failure model and effect analysis of industrial robots is fully utilized to connect the phenomena, causes and measures through the entities in order to build the industrial robot knowledge graph relationships. The constructed knowledge graph relationship is stored in a Neo4j graphical database, making it convenient for content retrieval and inquiry of application systems.In the industrial robot knowledge graph application side, the field maintenance personnel requirements are collected through a questionnaire survey and the requirements are classified into intent. A Bert+TextCNN structure model is built to realize the intention recognition of user inquiries. By combining entity recognition models and intent classification models, the system is able to better understand user inquiry needs, leading to the implementation of an intelligent maintenance system for industrial robots.