Diseinu Industrial Adimenduetarantz: Eredu Subrogatuen Bidezko Optimizazioa Testuinguru Errealetan

Surrogate based optimization was proposed 20 years ago to solve many real-world problems. The methodology has been used to solve problems in all areas of science and technology, and it has enabled the resolution of problems that could not be solved due to their computational cost or economic cost. However, although the methodology has been proposed to solve computationally complex or experimentally complex problems, research in the second case has been much scarcer.

Surrogate based optimization has become a well-known concept in recent decades, and although extensive research has been carried out in this field, there are still challenges and open questions: how to act in problems based solely on experimental data, examine the complexities that may arise in the substitution of complex physical phenomena (such as spatial, temporal or spatio-temporal phenomena), from the perspective of sampling, modeling and optimization algorithms. Furthermore, it is also essential to investigate the benefits that improvements in models can bring, by leveraging the opportunities offered by advances in machine learning.

The main hypothesis of this work is that surrogate based optimization is a useful tool for optimizing complex processes in industrial contexts, as long as more effective sampling methods are developed, appropriate models are chosen, and the optimization algorithms are tailored to the real context. In this regard, this thesis develops the research in three directions: firstly, working with simulated data, it is analyzed how adaptive sampling promotes efficient optimization; secondly, the ability of surrogate models to solve non-formalized problems through experimental data is highlighted; and, finally, the influence of research in machine learning on surrogate models is analyzed.

The scientific contributions of this work have been validated in two scenarios: (1) in the optimization of the production of rubber tyres in a Continental plant; (2) in the improvement of efficiency of the phase field model to accelerate the prediction of the process of solidification of metals in additive manufacturing.