In early 2025, the "Intelligent Prediction System for Steel Plate Convexity Based on Multi-model Fusion" was successfully launched on Baosteel Zhanjiang's 4,300 mm thick plate production line. The system is developed by MCC Capital Engineering & Research Incorporation Limited (hereinafter referred to as "CERI"). By integrating mechanism model and big data analysis technology, it realizes the intelligent upgrading of convexity control in thick plate rolling process, significantly improves product quality and production efficiency, and sets a benchmark for the intelligent transformation of China's iron and steel industry. 

Dual Drive of Mechanism and Data to Solve Complex Process Problems  

Convexity is the core quality indicator of thick plate products, which directly affects the market competitiveness of steel companies. In solving the problem of plate convexity control, traditional technical solutions rely on finite element simulation, which has problems such as low computational efficiency and poor adaptability. Although purely data-driven neural networks can learn input and output relationships through historical data, they have problems such as weak generalization ability. 

Anchoring the pain points and difficulties, the CERI project team came up with clever and practical measures. They innovatively proposed an intelligent framework of "dual drive of mechanism + data" to realize mechanism constraints and build a high-precision prediction model based on physical laws such as thermal expansion, wear and elastic deformation of rolls to ensure scientific calculation; realize data empowerment, use physical information neural network (PINN) to analyze production data in real time, dynamically correct model parameters, and break through the prediction bottleneck under complex working conditions; realize roll shape optimization, combine multi-objective genetic algorithms, solve the optimal roll shape curve under equipment constraints, and simultaneously improve convexity control accuracy and roll life. 

The system can predict the edge thinning and width bounce of steel plates in real time, and link the expert system to dynamically adjust the roll shape curve of the finishing mill, realizing a leap from "empirical control" to "intelligent decision-making". 

Focus on Production Capacity, Quality and Cost to Improve Application Results in Three Dimensions 

Knowledge comes from practices, and challenge shapes excellence. After production verification, the system has achieved significant benefits in Baosteel Zhanjiang's thick plate production line. 

In terms of quality upgrade, the average convexity deviation of steel plates is reduced by 23% to about 20 μm, the off-contract rate of steel rolling size is reduced from 0.45% to 0.23%, and the yield rate is increased by 0.02%; in terms of efficiency leap, the work roll change cycle of the finishing mill is extended by 16%, from 17.3 hours to 20.15 hours, the annual effective operation time is increased by 48 hours, and the production capacity is increased by 13 kt/a; in terms of economic benefits, the increase in profits per ton of steel and the improvement in yield rate have driven an annual benefit of more than RMB 6 million, of which direct profit increased by about RMB 4 million and quality optimization contributed more than RMB 2 million.