He/Him

Home Research Software Publication Teaching Contact

Motivation and Challenge

“Can we capitalize on the sizeable body of existing data to draw on past knowledge?” is the question to be answered in this postdoctoral research project. There are valuable lessons to be learned from the past. In this digital era, we have a vast amount of historical data to learn from, and machine-learning is the tool to transform implicit knowledge into an explicit form that engineers can tap into. However, it has also become much more challenging to implement conventional machine-learning models that can be effectively and efficiently trained using modern data that has become increasingly complex.

>40,000 boreholes across the entire Singapore from 1970s

Data-driven Three-dimensional Reconstruction of Subsurface Stratigraphy

Ministry of National Development Singapore

S$713,200 ($527,910)

2021-2023

In this research project, through an innovative hierarchical machine-learning implementation, the team revamped the conventional machine-learning implementation wherein a single AI model is built to learn from the data. In the revamped implementation, we first built many small-scale machine-learning models based on microscopic information. We then built a second-level machine-learning model that is trained using the results from the small-scale models, augmented with additional macroscopic information. We implemented the algorithm for engineering geology applications. Through close collaborations with government partners, we used the algorithm to reconstruct the three-dimensional geological interfaces delineating different geology units using real borehole data of ongoing underground constructions. We further showed that the algorithm can be seamlessly integrated into current geotechnical engineering practice, and it outperforms the state-of-the-art techniques that our government partner organization is using for ongoing projects. Notably, the software we developed is now being employed by our government partner for ongoing underground construction projects. With the backing of the Ministry of National Development of Singapore, we will initiate the process of commercializing the software.

Peer-reviewed Journal Articles

Wang, Z. Z., Hu, Y., Guo, X. F., He, X. G., Kek, Y. D., Ku, T.*, Goh, S. H., Leung, C. F. (2023), “Predicting geological interfaces using stacking ensemble learning with multi-scale features”. Canadian Geotechnical Journal. https://doi.org/10.1139/cgj-2022-0365. [PDF]

Software

Wang, Z. Z.*, Guo, X. F., Hu, Y., Goh, S. H., He, X. G., Leung, C. F., & Kek, Y. D. (2023). GeoMap: A Toolbox for Three-dimensional Reconstruction of Geological Stratigraphy [Computer software].