Bayesian optimization to infer parameters in viscoelasticity

Inferring viscoelasticity parameters is a key challenge that often leads to non-unique solutions when fitting rheological data. In this context, we propose a machine learning approach that utilizes Bayesian optimization for parameter inference during curve-fitting processes. To fit a viscoelastic model to rheological data, the Bayesian optimization maps the parameter values to a given error function. It then exploits the mapped space to identify parameter combinations that minimize the error. We compare the Bayesian optimization results to traditional fitting routines and demonstrate that our approach finds the fitting parameters in a less or similar number of iterations. Furthermore, it also creates a "white-box" and supervised framework for parameter estimation in linear viscoelasticity modeling.