Development of a Data-Driven Constitutive Model for Predicting Stress and Pore Water Pressure–Strain Relationships Using Soil Index Properties
| Student | Sarah Gustiarini Rifdah |
|---|---|
| Supervisor(s) | Dayu Apoji S.T., M.T., M.Sc., Ph.D |
| Year | 2025 |
Brief Abstract
Background
Geotechnical problems that are becoming increasingly complex demand a deeper understanding of soil behavior through constitutive modeling. To date, no constitutive model has been able to accurately represent the non-linear behavior of soil using only basic soil properties obtained from simple laboratory tests
Objective
This study aims to develop data-driven models of stress–strain and pore water pressure (PWP)–strain curves using Random Forest (RF), Multi-Output Random Forest (MORF), and Backpropagation Neural Network (BPNN) algorithms.
Methodology
With the advancement of Machine Learning technologies, data-driven approaches offer new opportunities to develop more efficient models of soil behavior. The models are developed using direct and incremental prediction approach, with soil index properties and variable histories as the input, without relying on additional correlations. The dataset used in this study was obtained from Consolidated Undrained (CU) triaxial tests on various soil samples.
Key Contribution
This study indicates that soil plasticity and strain level significantly affect prediction accuracy. Overall, this study highlights the potential of data-driven approaches as a simpler and more efficient alternative to constitutive models for future geotechnical engineering applications.