Monitoring and optimization of processes in the metallurgical industry through advanced multivariate statistical chemometric methods

The PhD project is devoted to the optimization of products/processes in the steel industry. A key innovation was the development of Total Quality Tutor (TQT), a software interface for integrating predictive models into real-time production processes. The workflow was developed in different steps: I) historical process data collection, synchronization and pre-processing; ii) model building for predicting the mechanical properties of steel products from the process parameters and the input material chemical characterization; ill) prediction of the optimal process parameters for the production of new steel products. All the steps are carried out by means of chemometric tools (pattern recognition, genetic algorithms, neural networks etc) supported by human reinforcement. TQT is designed for operating at all the levels of the plant in an automatic way integrating advanced data analytics and machine learning. Predictive models achieved exceptional accuracy in forecasting mechanical properties (yield strength, tensile strength, elongation), in the range 90-98% for elongation on low-carbon and interstitial-free steels. An important advance was the optimization of new production cycles, with success rates > 90% thanks to continuous refining. A new cycle launched in 2024 showed success rates > 99%. This highlighted their effectiveness in reducing costs while maintaining high-quality output. The results highlight the project's ability to optimize production and improve cost-effectiveness. The project also tackled steel aging, deepening the results obtained from a previous project, achieving significant insights into the parameters required to mitigate its effects, thereby improving material stability and durability. Over three years, these innovations yielded cumulative economic savings exceeding €3 million, emphasizing the transformative potential of machine learning in industrial workflows.