Enhancing 3D geological and geotechnical engineering model of Bangkok subsoil using optimal deep neural network models

Abstract

Understanding the geotechnical characteristics of subsoil is important for safety and efficiency in design and construction processes. In particular, the subsoil in the Bangkok Metropolitan area has accumulated soft marine clay over a long period, resulting in a thick layer of soft clay, which poses challenges for engineers. This study presents an approach to modelling the subsoil of the Bangkok Metropolitan region by utilising a large dataset of borehole data, enhanced with a Deep Neural Network (DNN) model, to develop a 3D geotechnical map. The hyperparameters of the DNN were tuned to fit the dataset for classifying the soil layers and the regression models were generated to predict the geotechnical engineering properties of the Bangkok subsoil, including the bulk unit weight, water content, plasticity index, undrained shear strength, and SPT-N values. The DNN model performance has been evaluated to ensure the accuracy and reliability of its predictions. The generated 3D geotechnical map was compared with the map obtained from the traditional kriging method to verify the map accuracy and differences in results between these two approaches. This study demonstrates the potential of machine learning techniques for improving geotechnical mapping and geotechnical engineering information. The outcomes of this research also support Sustainable Development Goals (SDGs), particularly SDG 9, by providing accurate geotechnical data to enhance sustainable infrastructure planning, and SDG 11, by refining the subsoil model in urban areas, which contributes to safer and more sustainable urban development while reducing environmental risks in construction.