Data-driven through-process modelling of aluminum extrusion: Predicting mechanical properties

This paper reports on a case study on zero-defect manufacturing, using a data driven approach for predicting final product properties from material and manufacturing process information. The case is a company that produces aluminum automotive parts, having an integrated value chain from casting, extrusion, forming and component assembly. This value chain is unique in terms of the intervened and interdependent system of material, process and product properties, where the traditional scientific approach has been to apply physics-based, through-process models to explain the complex thermo-mechanical evolution. Here, an alternative and complementary data-driven model is trained and tested on process data for a 6xxx-alloy product range using machine learning, which then is compared to a physics-based as well as a simple hybrid model. The presented results indicate that a machine learning approach is advantageous when numerous relevant observations are available, inherently adapting to the specific company-level conditions. The main strengths of a physics-based model, on the other hand, is general applicability and self-sufficiency with no need of prior observations. © 2024 The Author(s)