Hysteresis modeling of structural systems using physics-guided universal ordinary differential equations

In this research, we develop three hysteretic structural models using two machine learning algorithms, namely physics-guided neural networks and universal ordinary differential equations. These models employ multilayer perceptrons and laws of nature to identify the hysteretic behavior of structures in an unbiased and physically consistent manner. These models might be regarded as an extension of the Bouc model of hysteresis, that enforces the Drucker and Il'iushin postulates in addition to other physical principles. We validate the proposed framework with experimental data of ferrocement and recycled plastic lumber walls, obtaining good accuracy, generalization and physical consistency. All data and codes used in this work are publicly available on GitHub.(c) 2023 Elsevier Ltd. All rights reserved.