Abstract

In the past few decades, the demand of construction in underground spaces has increased dramatically in urban areas with high population densities. However, the impact of the construction of underground structures in surrounding infrastructures raises a lot of concerns since the movements caused by deep excavations might damage adjacent structures. Unfortunately, the prediction of the geotechnical behaviour is difficult due to uncertainties and lack of information of the underground environment.Therefore, to ensure safety, engineers tend to choose unfavourable conditions for the design of excavation supporting systems, which usually leads to a conservative design that requires unnecessary material and construction time.

Adaptive design provides a way to avoid such redundancy by using the most probable conditions and incorporating knowledge learned during the construction progress. The monitoring data obtained during the construction is used to update the model and therefore the accuracy of the prediction of the ground response will be improved. This process is recognised as the back analysis,a core procedure in the Observational Method.

This process can be realised by using supervised learning. In this research, a probabilistic model coupled with Bayesian inference is developed which is not only able to learn the relations between the input soil parameters and the response, but also identify the underlying uncertainties from all sources. Moreover, it integrates subjective information and objective engineering experience information in a rational and quantitative way. Furthermore, under this probabilistic setting, the uncertainty information is also contained in the prediction, which is crucial to the confidence based decision making.