Surrogate Modeling Framework for Cantilever Sheet Pile Wall Stability Assessment: Case Study Campus Ullevål

This study presents a surrogate modeling framework for evaluating the stability of cantilever sheet pile walls in soft clay deposits and applies it to the real-world excavation project Campus Ullev˚al. Key design variables and soil parameters driving cantilever sheet pile wall stability are varied to generate a representative dataset for training surrogate models based on strength-reduction finite element analyses. The surrogate models are constructed using polynomial chaos expansions to provide fast and accurate factor of safety predictions across a wide design variable space, avoiding repeated solution of computationally expensive boundary value problems. Sensitivity analyses confirm that the undrained shear strength of the clay layers is the dominant factor, whereas the properties of the lime-cement stabilization have a subordinate effect in the reference case. The framework further enables fully probabilistic analyses by quantifying uncertainty propagation and reliability metrics, showing that reducing epistemic uncertainty may be essential to meet target reliability requirements. This underscores the importance of thorough site investigation and provides a transparent basis for communicating its value to project stakeholders. Beyond sensitivity and reliability analyses, the surrogate modeling framework can be extended to support real-time back-analysis of model parameters and automated design optimization.