Mapping shear strength and compressibility of soft soils with artificial neural networks

Soft soils are widely distributed in the Guangdong-Hong Kong-Macao Greater Bay Area of China. The soft soils are featured with large water content, high compressibility and low permeability, posing great challenges in dealing with bearing capacity and foundation settlements. Extensive laboratory tests have to be conducted to determine parameters for shear strength and compressibility properties of soft soils. This is really time consuming and costly. In addition, sample disturbance and lab testing error are also unavoidable. The aim of this study is to develop a simple, efficient while satisfactorily accurate tool for prompt assessments of parameters for shear strength and compressibility of soft soils. The artificial neural network (ANN) technique is employed to reach this goal. A large database is first presented for measured physical and mechanical parameters of soft soils sampled from a core city in the Greater Bay Area. The data are obtained from six types of geotechnical laboratory tests, including direct shear test, consolidation test, unconsolidated undrained test, total stress consolidated undrained test, effective stress consolidated undrained test and compression test. Then, the ANN is applied to map the shear strength and compressibility properties of the soft soil from its physical parameters. The analytical forms of the ANN models are derived and presented to enhance their practical value. Next, the accuracies of the six ANN models are evaluated using model bias statistics where model bias is the ratio of measured to predicted value. The evaluation results showed that the ANN models are practically unbiased on average and the dispersions in prediction accuracy are low. Furthermore, the probability distributions of the model biases are characterized. This study helps saving time and cost of geotechnical investigation for soft soils in the area.

Automated summary

This study develops a simple, efficient, and accurate tool for prompt assessments of shear strength and compressibility parameters of soft soils in the Guangdong-Hong Kong-Macao Greater Bay Area of China. By employing the artificial neural network (ANN) technique, the study maps the shear strength and compressibility properties of soft soils based on their physical parameters and evaluates the accuracy of the ANN models. The study helps save time and cost in geotechnical investigation for soft soils in the area.