Research On The Construction Of Industrial Robot Prognostic And Health Management Knowledge Graph

The prognostics and health management(PHM)technology of industrial robots obtains the health or failure status of the robot in real time by analyzing the operating data and status data of the robot,and predicts whether a specific part of the robot will fail or has failed.,So as to take certain measures to solve the fault that is about to occur or have occurred,thereby improving the efficiency of robot work and production,and reducing the cost of maintenance and repair.In recent years,natural language technology and knowledge graph technology have developed rapidly.These technologies have brought the possibility of solving problems such as the related expression of data and knowledge in the field of intelligent manufacturing and knowledge reuse,and they have played an increasingly important role in the process of intelligent manufacturing..Therefore,construct an industrial robot PHM knowledge graph with complete knowledge coverage and high knowledge quality,realize the knowledge representation,knowledge sharing and reuse of industrial robot PHM information,and provide intelligent semantic search,auxiliary decision support and other functions for machine maintenance personnel.Very important practical significance.Based on the investigation and analysis of existing knowledge graph construction and application methods,this paper studies the PHM knowledge graph construction and application methods of industrial robots,and proposes the three steps of the knowledge graph construction: ontology construction,crowdsourcing semantic annotation,and information extraction.The main thesis is Contributions include:(1)A method for semi-automatic construction of the PHM ontology of an industrial robot is proposed.On the basis of ensuring the completeness of the construction with a variety of data sources,a comprehensive algorithm integrating word frequency,document frequency,TF-IDF algorithm,and C-value algorithm is used to Carry out ontology concept extraction,in the relationship recognition stage,measure the correlation strength of concept pairs through Dice measurement,mine subordinate relationships based on the combination of CSC semantic lexicon and search engine,and identify cross relationships between concepts based on SAO structure,and finally use Protégé software to form Ontology.(2)Explore the iterative mode of "ontology construction" and "semantic annotation",use the constructed ontology to guide the annotation of unstructured text,obtain the training data set required for "information extraction",and update the knowledge of the ontology.The entity relationship joint extraction method combining Bi-LSTM-CRF and multi-head selection is applied.The model shares the coding layer and can perform named entity recognition and relationship extraction in information extraction at the same time.The multi-head selection structure in the model can effectively solve the relationship.Overlapping problems,experiments show that this model can effectively extract robot PHM knowledge.(3)Discuss the application mode and application scenarios of the application platform based on the industrial robot PHM knowledge map,design and implement the industrial robot PHM knowledge map application platform.