Development of Machine Learning Methods for Planning of Geotechnical Site Investigation and Analytics of Site Investigation Data

Description

Site investigation is indispensable to geotechnical practice. Interpretation of site investigation data leads to the expected ground conditions (e.g., spatial distribution of different soil/rock types) and geotechnical property design profiles that can then be used in geotechnical analyses and designs. Without proper site investigation or data analytics to produce reliable input parameters, the subsequent geotechnical analyses and designs are much less meaningful, and the construction projects might be subjected to significant risk, such as project delay and over-budget caused by an “unexpected site condition” (i.e., the actual ground conditions encountered during construction differ significantly and adversely from the site investigation results). For example, the “unexpected site condition” in the Guangzhou-Shenzhen-Hong Kong Express Rail Link project commissioned in 2018 caused a project delay of more than 2 years and an over-budget of HK$20 billion. Although the importance of site investigation is well-recognized, NO scientific or quantitative method currently exists for site investigation planning, including determination of the optimal locations and necessary number of investigation points (e.g., boreholes or cone penetration tests (CPTs)), number of specimens to be tested, and the locations and depths at which specimens shall be taken. Existing design codes around the world provide only conceptual principles for such planning. This project aims to develop machine learning methods that leverage on the multi-stage nature of site investigation and use innovative data analytic methods (e.g., compressive sensing (CS)) to analyze data obtained from previous stages (e.g., desk study or preliminary stage) and automatically make recommendations on planning of the current investigation stage (e.g., the optimal locations and necessary number of investigation points or specimens to be tested). The PI has performed preliminary studies for one- dimensional cases and obtained promising results. Major research components of the project include: (1) development of entropy-based methods to determine the optimal sampling locations, (2) development of CS-based methods to determine the necessary sample number, (3) calibration of the proposed methods using simulated and real data, and (4) development of computer software to facilitate application of the proposed methods. For the first time ever in geotechnical engineering, this project will provide quantitative and data-driven machine learning methods to facilitate informed decision-making in site investigation planning. It will also offer computer software to facilitate application of the proposed methods in engineering practices. The research outcomes will be beneficial to geotechnical practice worldwide and in Hong Kong specifically, where many large-scale infrastructure projects are currently in progress.