Machine learning driven prediction of mechanical properties of rolled aluminum and development of an in-situ quality control method based on electrical resistivity measurement

As a result of the increasing digitization in the context of the fourth industrial revolution in the metal processing industry, the application of the associated possibilities has also entered the field of optimized in-situ material quality testing. In-line monitoring in the production of semi-finished products such as sheet metal in particular will be implemented and optimized to an increasing extent. This study introduces a novel approach that establishes a correlation between the resultant mechanical properties, characterized by stress-strain values and related parameters such as the n-value, with the electrical resistance of aluminum sheets for a rolling process. This correlation is examined under varying rolling schedules and lubrication conditions, utilizing a non-contact measurement system for data acquisition. The four-point method, which was implemented for material testing on a tensile testing machine, can be further transferred to a rolling aggregate for an in-situ quality control step between the passes. Thus, using a black box machine learning approach, conclusions can be drawn about the prevailing mechanical properties as well as the anisotropic behavior by measuring the electrical resistance. Furthermore, a graphical user interface was developed, which generates an optimized rolling schedule for desired mechanical properties. By integrating such a measurement equipment into the rolling process, predictions can be made about the resulting properties and labor-intensive subsequent quality checks as well as scrap can be significantly reduced. In addition, there is the possibility to specify desired mechanical properties for further sheet forming processes, whereupon the optimal rolling process route to achieve them is displayed.