A knowledge graph construction and causal structure mining approach for non-stationary manufacturing systems

Knowledge graph (KG) is a method for managing multi-source heterogeneous data and forming knowledge for reasoning using graph structure. It has been extensively utilized in manufacturing systems to promote the advancement of intelligent manufacturing. In non-stationary manufacturing systems, the machining performance of individual elements demonstrates variability and dynamic fluctuations. The significant dynamics and uncertainties of a manufacturing system bring great challenges to KG's modeling, construction, and reasoning. To overcome these challenges, this paper proposes a Digital-Physical Manufacturing Knowledge Graph (DPMKG) construction and reasoning method. Firstly, an ontology-based knowledge representation model is developed to facilitate the integration of digital domain knowledge with the description of physical domain performance fluctuations, thereby establishing the schema layer of DPMKG. Secondly, a SysML model-driven construction pipeline is proposed to facilitate the correlation and integration of multi-source data from both digital and physical domains, thereby establishing the instance layer of DPMKG. Thirdly, a causal structure mining method for DPMKG is developed to enhance the analytical and reasoning capabilities in non-stationary manufacturing systems. Finally, an aero-engine casing machining system is employed as a case study to establish the DPMKG, and reasoning is performed on the process quality prediction task. The case study reveals that the proposed DPMKG modeling, construction, and reasoning approach can effectively describe and analyze performance fluctuations in the physical domain of a non-stationary manufacturing system. By integrating digital and physical domain knowledge, the extensive data can be effectively leveraged to generate knowledge for reasoning, thereby facilitating intelligent and refined control of non-stationary manufacturing systems.