Using data-science models to predict technological factors affecting the mechanical properties of flat products

In recent years, the stringent specifications lead to the need of more strict regulations within the plants in order to provide products that conform to the market needs. This requires a better appraisal of the mechanisms that influence the demanded material properties in order to set up the appropriate rules. For the production of steel plates for example, a complete steelmaking route from scrap melting, refining, casting, till rolling is required. A great deal of information linked to technological parameters involved in the process is already stored in the automation systems within a plant. With the deployment of machine learning algorithms specific mechanical properties of the final products can be related with these factors and two major results can be obtained:The most important technological parameters that influence the properties under investigation are deduced;A supervised model that predicts a mechanical property upon a set of input data can be derived within a measurable statistical error.In this work, two mechanical properties for produced plates, the tensile strength (Rm), and the yield stress (Re), were analyzed with respect to 33 independent parameters representing salient features in the whole production process. Three data science models, the deep learning, the distributed random forest, and the gradient boosting method were deployed for securing the validation of the results. Attention is drawn upon those technological parameters that are top selected in all three models. Actual and predicted properties values are also presented.Keywords: data science models, technological parameters, flat products, predicted properties.