Project R-15075

Data-efficient Machine learning for Smart Design of Experiments (Research)

In many industries, traditional manufacturing processes are replaced by novel technologies, e.g., because products need to meet ever stricter environmental concerns. The parameters of these novel processes may heavily influence the characteristics of the products produced (economical, environmental, technical, etc.). The challenge, then, is to detect process settings that optimize the intended performance in a cost-efficient way. Adhesive bonding is a technology that is increasingly relevant in settings where traditional methods (like welding or clinching) are no longer viable. It requires careful parameter selection, through resource-intensive trial/error experiments. A sequential Design of Experiments (DoE) using machine learning can streamline this optimization, increasing both data efficiency and solution quality. This project aims to close the gap between the Bayesian optimization methods for smart Design of Experiments (DoE) developed in GC1 (i.e., Task 4.4 and the JMLab PoC) and the needs of industrial companies to handle a larger variety of use cases, thereby increasing the Technology Readiness Level from 3 to 4.

01 May 2024 - 31 August 2024