Christoforos Rekatsina, Panagiotis Krokidas, Vasileios Vavourakis, Clara Essmann, George Giannakopoulos
Abstract
Predicting material properties, such as Young’s modulus, in complex multilayered-multimaterial structures is a challenging task that requires the integration of data-driven methods with physical principles. In this work, we present a Physics-guided Neural Network framework for estimating multiple Young’s moduli in multimaterial cylindrical structures, with a specific application to Caenorhabditis elegans worms. Our approach leverages contact mechanics to model the force-indentation behavior of multilayered ring-cylinder systems, providing a physics-based foundation for the neural network. To ensure physically consistent predictions, we introduce a custom activation function that enforces bounds on the estimated Young’s modulus using hyperbolic tangent transformations and scaling. The neural network architecture is designed to handle the complexity of multilayered systems, and the training process incorporates progressive tightening of bounds to balance exploration and exploitation. We validate our framework using both numerical data from high-fidelity finite element models and experimental force-indentation measurements from C. elegans worms, demonstrating the framework’s ability to accurately predict Young’s modulus for each layer of the cylindrical structure. This work highlights the potential of combining physics-guided machine learning with domain-specific knowledge to solve complex inverse problems in material science and biomechanics, offering a versatile and efficient alternative to state-of-the-art global optimization techniques like the Bayesian optimization algorithm.