Dalarna University's logo and link to the university's website

The steel industry involves a series of intricate processes, from the mining of iron ore to the production of finished steel products. Among these processes, the production of cut-to-length (CTL) sheets is vital for transforming steel coils into flat sheets tailored to customer specifications. This study focuses on optimizing the flattening process within the CTL line, specifically targeting the machine settings of two levelers, working in a series, which play a critical role in ensuring sheet flatness. The flattening process is currently reliant on predefined machine settings and operator expertise, often leading to inefficiencies, increased scrap rates, and the need for reprocessing.

This research leverages historical production data and employs advanced statistical techniques to address these challenges. Using correlation analysis and regression modelling, the study identifies key variables affecting flatness deviation. Response surface methodology (RSM) is then utilized to derive optimal machine settings that minimize these deviations. The performance of RSM implied optimal settings was evaluated using a validation data set.

Among the models tested, Ridge regression demonstrated superior performance with ahigher R², and lower mean squared error compared to ordinary least squares regression. The validation study further confirmed the effectiveness of the optimized settings, with minimal Euclidean distance between actual and predicted machine settings, highlighting the practical applicability of the findings.

Future improvements include narrowing the settings range around the identified optima, conducting real-time production experiments, and expanding the dataset to include more comprehensive material classifications. Simulation-based data augmentation is also proposed to enable controlled experiments for specific material groups.

The proposed standardized settings reduce the dependency on individually unique adjustments for different material groups, providing a consistent and efficient alternative. The outcomes of this research offer a foundation for future data-driven optimization in steelsheet leveling, with practical implications for reducing scrap rates and improving production quality. The training dataset was compiled from a sample of various material groups, which limited the scope of analysis for specific material types.