Contribution
Statistical Assessment of a Gradient-Based Optimization Method for Material Parameter Determination of Polymers
* Presenting author
Abstract:
This contribution addresses the identification of frequency-dependent elastic and damping parameters of polymers in the ultrasound range formulated as an inverse problem based on waveguide simulations. Specifically, the polymers of interest are PEEK, PA6, and PP, focusing on hollow cylinder geometries to enhance simulation efficiency.We examine the impact of both measurement setup and initial parameter estimates on the convergence behavior of gradient-based optimization methods. To accelerate optimization, we introduce a novel adaptation of the Levenberg-Marquardt method and an improved objective function based on the cross-correlated envelope of signals. Given that this study primarily relies on simulation data to quantify convergence, we research expected parameter ranges and analyze correlations between anisotropy and damping characteristics for the materials of interest. Based on these insights, we define distributions for the elastic parameters and draw samples for our primary study to extrapolate results to measurements.Our findings demonstrate that optimization divergence is frequently observed when an excitation is applied uniformly across the cylinder’s end face. In contrast, non-uniform excitations significantly increase optimization robustness, supporting prior assessments and highlighting the potential of meta-optimization techniques, such as Bayesian optimization, for further improvement via modification of sample and excitation geometry.