A method for constructing a machining knowledge graph using an improved transformer

Abstract

Process knowledge base is a core component in the intelligent process, which determines the intelligent degree of product manufacturing and directly affects the production efficiency of products. However, traditional process knowledge base is often constructed manually, which is difficult and time-consuming. In addition, in the field of machining, there is a large amount of unstructured invisible process knowledge, which is not effectively organized and managed. To make use of this knowledge and provide knowledge support for downstream production and maintenance, a process knowledge base construction framework is proposed by using Knowledge Graph (KG) technology. Firstly, the ontology rules of process knowledge are designed from the perspective of the processing method of process characteristics according to the particularity of knowledge in the machining field. The process KG schema layer is then constructed. Secondly, a neural network BERT–Improved TRANSFORMER–CRF (BITC) model is proposed for the machining knowledge extraction task, and the data layer is constructed. Then, entity linking and knowledge fusion are performed by using the word vector cosine similarity algorithm and stored in Neo4j. The process KG is then constructed. Finally, the effectiveness of the proposed method is verified by using an aero-engine casing of an enterprise as an example. Under the same dataset, the BITC model scored higher than several other classical models. The Precision, Recall, and F1-score were 85.27%, 86.40%, and 85.83 %, respectively.