Accelerated identification of hardening soil model parameters employing tree-based machine learning algorithms

Abstract This paper proposes an approach for identifying the constitutive model parameters of an elasto-plastic double hardening model using systematically optimized, tree-based supervised machine learning algorithms. The machine learning model utilizes data from two laboratory tests—a drained triaxial test and an oedometer test with an unloading-reloading cycle—to predict the constitutive model parameters. The accuracy of the machine learning model is demonstrated through three comparison approaches: (i) the parameter optimization tool implemented in a commercial finite element software applied to a synthetic dataset representing sand, (ii) parameter optimization tool and experience-based parameter determination with a real dataset of Karlsruhe fine sand. The experience-based determination survey involved independent experts who calibrated the model parameters based on the same input data used by the machine learning model. The machine learning model predictions align closely with the target parameters and the results from the parameter optimization tool for synthetic tests, and the developed approach matches the parameter ranges obtained from both automated identification and the experience-based survey for Karlsruhe fine sand. The model effectively captures the triaxial test results and the oedometer test results, outperforming the average experience-based results. In terms of computational efficiency, the proposed machine learning approach reduces the average determination time by approximately 19 500 times compared to manual parameter identification and by about 2 500 times compared to automated identification. The proposed machine learning approach is considered superior due to its speed, consistency, and scalability compared to both experience-based and automated identification methods. This approach can be readily adapted to other constitutive models, promoting the use of more advanced models in practice.