Metallurgical Data Science for Steel Industry: A Case Study on Basic Oxygen Furnace

Abstract

Steel production is one of the biggest and most important industries in the world outputting hundreds of tons of steel daily. A steelmaking plant pushes the conventional methods of process monitoring and control to their limits due to the complexity and multidimensionality involved in the physical, mechanical, and chemical metallurgy. The manufacturing process of steel plates involves multiple steps such as blast furnace smelting, converter smelting, and ladle furnace refining, followed by continuous casting, heating, rolling, and cooling. Each physical process generates numerous "key process variables" such as steel composition, additives, environmental control, cooling, and other process parameters, all influencing each other, the subsequent processing steps, and hence the final product. Therefore, modeling and digitally twinning of such processes and predicting the quality of steel through comprehensive finite element approach (FEA) calculations or experimental trials is time-consuming, costly, and impractical. In recent years, this complexity, the increasing global competition, and the drive for more efficient lower waste production created a high demand for new methods to optimize the steel production processes and the mechanical properties of final products. [1] In such a technology-intensive industry, even the smallest variation during the production process causes costly and time-consuming postprocessing or an increase in scrappage. [2] Therefore, smart, agile data-driven prediction models are necessary and urgently needed. To satisfy the demanding requirements for "Industry

Dynamic Prediction with Statistical Uncertainty Evaluation of Phosphorus Content Based on Functional Relevance Vector Machine

“…Meanwhile, the sampling process would break the continuity of the production process and decrease production efficiency. By contrast, the model prediction way [1,2] provides a soft measurement way to estimate the molten steel's phosphorus content, such as first-principlebased models, knowledge-based models, and data-based models. As the black-box state of the steelmaking process makes the real-time metallurgical reaction state unknown, and the continuously innovated technology as well as complex production process make knowledge accumulation difficult, the data-based models develop rapidly.…”

End‐Point Prediction of Converter Steelmaking Based on Main Process Data

“…These data‐driven models outperform traditional mathematical and dynamic detection methods in accuracy. [ 7,8 ] In addition, these models are not limited by the conditions of production equipment. Consequently, machine‐learning technologies for end‐point prediction in converters hold considerable promise.…”

Control of the Molten Steel Level in the Top Side‐Pouring Twin‐Roll Casting Process Based on Fuzzy Rules Optimized by Particle Swarm Optimization Algorithm

“…[18,22,26] At present, popular artificial intelligence (AI) methods including traditional statistical methods, traditional symbolic AI, and computational intelligence (CI) are applied to solving complicated real-world issues for which conventional approaches are insufficient or impracticable. [27][28][29][30][31] Artificial neural networks (ANNs), fuzzy logic (FL), and evolutionary algorithms (EAs) are the most frequently employed CI techniques to solve different engineering problems. [27,31] EAs is a class of global optimization methods with wide applicability, [27] which provides a new way to optimize the fuzzy rules proposed by ordinary operators.…”