Abstract: The rolling process is characterized by strong nonlinearity, multivariable coupling, and variable operating conditions. Traditional mechanism models are difficult to accurately characterize the true state of rolling, while simple data-driven methods lack generalization ability and the decision-making process is difficult to trace. In the background of the technological transition from "smart manufacturing" to "artificial intelligence+" in the manufacturing industry, how to systematically embed industrial mechanisms into AI models has become a core issue that urgently needs to be broken through. This article systematically summarizes three main modes of deep integration between industrial mechanisms and AI from a methodological perspective, namely mechanism and data hybrid modeling, mechanism guided data modeling, and semantic fusion data modeling. On this basis, combined with typical scenarios such as properties prediction, precise control, quality traceability, equipment operation and maintenance, logistics and scheduling, relevant application solutions are presented,which provides a valuable reference for the development of intelligence rolling technology.