Evaluating the impact of geotechnical uncertainty in deep excavations through probabilistic design

Abstract

As geotechnical projects increase in complexity, such as deep excavations with embedded sheet pile walls in soft soil, reducing uncertainty and ensuring reliable design become challenging. Misunderstanding the inherent variability of soil properties, such as undrained shear strength and stiffness, can lead to the misestimation of safety margins and structural performance. While traditional hand calculations using partial factors can address uncertainty, they often fall short in complex projects involving different failure mechanisms. In contrast, probabilistic analysis combined with finite element methods offers a more applicable approach for estimating credible failure probabilities and ensuring the reliability of geotechnical design. The primary purpose of this thesis is to apply a fully probabilistic design approach under a reliability index of three, considering two different coefficients of variation (COV) of 10% and 20%. Additionally, the study investigates the impact of variability in two key clay properties and quantifies the performance of retaining walls with their lateral support components. Lastly, it evaluates the limitations and strengths of the probabilistic design approach compared to traditional hand calculation methods. To demonstrate the benefits of a fully probabilistic design, 2000 random calculations were generated using Monte Carlo simulation in PLAXIS 2D. All generated results were then analyzed and compared with deterministic designs. The findings indicate that applying 20% COVs increased the predicted range of the wall’s maximum bending moment, waler force, and horizontal deflection, showing a broader response. The significant effect of traffic loads combined with higher soil variability resulted in high wall deflection and lower factors of safety in the majority of realizations. The convergence of deterministic and probabilistic designs underscores the importance of applying probabilistic methods to achieve more reliable and uncertainty-informed geotechnical designs.