Rubber is used in various industrial products and has become an indispensable material in modern society. In the constitutive laws of rubber, low-order models are insufficiently accurate, while the pursuit of high accuracy results in models containing high-order polynomials, which not only makes physical interpretation difficult but also increases the cost of parameter identification from experimental data. In this study, we construct a constitutive law for rubber using machine learning for the purpose of seamless use of constitutive laws from experimental results. Data is generated using the simulation software Marc, which is then used as training data for machine learning. Specifically, we will construct two types of predictors. Predictor 1, which takes stress and strain as inputs and outputs equivalent C01 and C10 with reference to the Mooney model, and Predictor 2, which takes C01, C10, elongation ratio, and strain invariants as inputs and predicts strain energy W, to achieve both accuracy and interpretability. The effectiveness of the proposed method is discussed.
HARASAKA et al. (Wed,) studied this question.