DATA DRIVEN MODELLING OF SOIL STRESS STRAIN BEHAVIOR USING MACHINE LEARNING

• Main Author: Gustiarini Rifdah, Sarah
• Format: Final Project
• Language: Indonesia
• Subjects: Teknik sipil
• Online Access: https://digilib.itb.ac.id/gdl/view/82166
• Institution: Institut Teknologi Bandung

Geotechnical problems have become more complicated, requiring the simulation of soil behavior at the element level to achieve awurate solutions. Constitutive models are used to simulate such complicated complex nonlinear soil behavior. However, the development of soil constitutive models results in increasingly complex models with unintuitive input parameters that often lack physical meaning and require correlation or assumption to determine. To date, there is no constitutive model capable of awurately representing the complex nonlinear behavior of soil using fundamental soil properties directly measured from laboratory tests. Machine learning has opened new opportunities to solve this problem as a tool to identify patterns between the fundamental soil properties and its stress-strain behavior. This study aims to develop a data-driven constitutive model that is capable of representing the complexity of soil stress-strain behavior using only its fundamental soil properties obtained from laboratory tests. This data-driven constitutive model is developed based on a database comprising numerous soil stress-strain behavior datasets obtained from triaxial test data and their corresponding soil properties collected from various locations across Indonesia. Two machine learning algorithms were employed: Ordinary Least Squares (OLS), representing linear models, and Random forests, representing nonlinear models. The results of this study indicate that the machine learning-based constitutive model, particularly the nonlinear model, can directly capture the nonlinearity of soil stress-strain behavior using only its fundamental soil properties with reasonable awuracy. This study highlights the potential of data-driven constitutive modeling to play a significant role in modeling soil behavior for geotechnical applications in the future