Mechanism data-driven modeling of stochastic wear and degradation of rolls in hot finishing mill

Abstract

During hot rolling process, the rolls are highly susceptible to wear and degradation, which directly affects the shape and surface quality of the strips. To reflect the wear and degradation behavior of rolls in actual production and to provide a scientific basis for decision-making in their management and maintenance strategies, a research approach integrating mechanistic and data-driven methodologies was employed in this study, considering the stochastic nature of the degradation process. Firstly, a single-task roll wear mechanism model grounded in tribology principles was developed. Subsequently, a multi-task wear mechanism model for a single rolling unit was formulated. Concurrently, a stochastic residual model was introduced to encapsulate the uncertainties inherent in the wear mechanism model. Utilizing actual measured data from a specific steel plant, the parameters for both the mechanistic and stochastic wear models were estimated, their effectiveness and precision were confirmed with an independent dataset, and a comparative analysis was conducted against industry-standard models. The findings reveal that the proposed model outperforms traditional models in predicting roll wear and degradation trends with enhanced accuracy, offering substantial practical application benefits. The integration of the stochastic model establishes a theoretical groundwork for reliability modeling and optimized roll management decision-making.