Adaptive prediction of wall movement during excavation using Bayesian inference

In underground construction works, uncertainties and insufficient information about the underground environment lead to inaccurate predictions of soil-structure interactions. Supported excavations are often over-designed, which underscores a significant potential for cost optimization. However, the uncertainties exist, and the traditional design process does not allow for leaner designs at the start of the project. The emergence of advanced analysis tools enables the development of an Observational Method based approach for a decision-making process in which data can be best utilized to deliver real value, confidence, and control.An automated back analysis approach based on Bayesian inference is developed in this paper and validated with a synthetic case study. Probabilistic modeling and Markov Chain Monte Carlo simulation are used to deliver estimates of soil parameters for a given a geotechnical model, update the prediction of future excavation stages, and fully quantify uncertainties from the constructed model and measurements. Sensitivity analysis is used for model selection to achieve modeling robustness. The impact of prior engineering knowledge about the soil properties on the precision of the predictions is also examined. This approach significantly improves the efficiency of back analysis in current practice and provides a tool for data-driven decision making of design optimization during construction.