Enhancing Reliability of Steel Production Lines Through Advanced Knowledge Graph-Based Fault Diagnosis Model

Reliability in steel production lines is critical for maintaining operational efficiency, ensuring product quality, and minimizing downtime. The potential failure of any equipment can negatively impact the reliability of steel production lines. Therefore, it is urgent to establish an efficient system for fault diagnosis (FD). We introduce a novel FD model called triple agent-based reinforcement learning (TARL), which comprises the fact extraction (FE) agent, the relation selection (RS) agent, and the entity selection (ES) agent. A top–down construction method is first employed to build a steel production line fault knowledge graph (SteelFaultKG). During the fault reasoning process, the RS agent and ES agent collaborate to identify the correct reasoning path on the SteelFaultKG. However, the initial SteelFaultKG is sparse and incomplete, hindering the RS and ES agents from inferring the correct reasoning path. To mitigate this limitation, the FE agent is introduced to dynamically enrich the SteelFaultKG by generating fact triples from an external corpus. This expansion of action space in SteelFaultKG is instrumental in enhancing the model's reasoning capabilities. Comprehensive experimental results on a steel production line fault-related dataset indicate that the TARL model significantly enhances FD accuracy. This advancement in FD, offered by the TARL model, significantly enhances the operational reliability of steel production lines.