Knowledge Graphs and Intelligent Engineering Systems: Theoretical Integration and Applications in Industrial Decision-Making

Abstract

Knowledge graph is a structured representation of domain entities, their attributes and their relationships, coded as a labeled directed graph. They have become the main way to integrate different and scattered knowledge into intelligent engineering systems. Different from the statistical machine learning model in which knowledge is hidden in its parameters, knowledge graph shows the reasons, classifications and process relationships in a way that can be understood by machine and human experts. This is why they are very good at being the semantic basis of industrial decision-making. This review focuses on knowledge graphics technology and how to apply them to intelligent engineering systems using network physical system (CPS), system engineering and information theory. She said that the knowledge map is a semantic compatibility layer, which allows the representation of statistical learning to be connected with the ontology in the engineering field. We analyze four transformation paths that artificial intelligence enhanced by knowledge graph promotes the development of manufacturing industry: from production knowledge enclosed in isolated data islands to semantically integrated knowledge, from reactive fault response to ontology-driven predictive diagnosis, from isolated machine control to knowledge-based autonomous operation, from independent subsystems to unified network physical ecosystem based on ontology. We have studied the application of fault detection based on knowledge graph-guided reasoning, knowledge-based digital twin manufacturing cell, intelligent process optimization based on ontology constraint learning and man-machine collaborative decision support. We use industry use cases and written frameworks. We have studied some thorny problems, such as the bottleneck of knowledge acquisition, the decay of temporary knowledge, and extensible reasoning under incomplete graphics. We also discussed the future things, such as automatic knowledge graph construction, neural symbol integration and cognitive digital twins. A new idea is to regard the completeness of knowledge graph as a measurable thing, and it depends on each application, rather than a perfect goal that can never be achieved.