A Hybrid Process Data and Knowledge-Based Fault Diagnosis Method for Product Quality Monitoring of a Hot Rolling Mill

In the hot tandem rolling mill (HTRM) process, product quality such as the strip exit thickness, crown, and temperature are often affected by different process faults. Thus, it has become a key step for process engineers to develop fault diagnosis (FD) methods, to detect potential faults and identify their probable root causes. In this paper, a hybrid process data and knowledge-based FD method is proposed and applied to a real HTRM process. A distributed canonical correlation analysis (CCA)-based fault detection method is first developed to accomplish: (1) multiple stands connected using the network can transmit the compressed quality-relevant information to the quality server; (2) a real-time fault detection module can be performed on the quality server to report a fault. To this end, the faulty stands affected can be roughly located through the contribution analysis-based method. Then a process knowledge based signed directed graph (SDG) is constructed, and based on this, the SDG reasoning-based method is developed to find the root causes. Finally, the feasibility of the proposed method is demonstrated using actual HTRM data.