Location: Room T4 at Aalto CS building + zoom

Abstract: Design of experiments and response surface methodology are considered as the gold standard for optimization in industrial experiments. The principles for describing response surfaces were originally established by chemists and statisticians and have largely remained the same since the 1950s. These traditional methods have recently been complemented with machine learning alternatives based on Bayesian optimization. Bayesian optimization combines experimental design and optimization into a single process and can potentially reduce the number of experiments required to reach optimal process conditions.

In this talk I will introduce Bayesian optimization from the perspective of design of experiments. I will focus on simulations and experiments for comparing these two methodologies in optimizing empirical systems where exact physical or chemical equations either do not exist or do not work properly. Finally, I will discuss the benefits and drawbacks of Bayesian optimization for design of experiments and introduce some narratives for its application.

References:

Rummukainen H., Hörhammer H., Kuusela P., Kilpi J., Sirviö J., Mäkelä M., Traditional or adaptive design of experiments? A pilot-scale comparison on alkaline wood delignification, Heliyon (2024) e24484 (10.1016/j.heliyon.2024.e24484)

Mäkelä M., Experimental design and response surface methodology in energy applications: a tutorial review, Energy Conversion and Management 151 (2017) 630-640 (10.1016/j.enconman.2017.09.021)

Bio: Mikko Mäkelä is a Research Professor in Intelligent Biomass Processing at VTT Technical Research Centre of Finland Ltd. He coordinates the AI-driven design of materials highlight at FCAI together with Milica Todorovic. His research focuses on empirical measurements and data-driven mathematical models for material and process development with an interdisciplinary background in environmental and process engineering.