Machine learning-based uncertainty modelling of mechanical properties of soft clays relating to time-dependent behavior and its application

Uncertainty is a commonplace and significant issue in geotechnical engineering. Unlike conventional statistical and machine learning methods, this study presents a novel approach to correlating soil properties that takes uncertainty into account using an artificial neural network with Monte Carlo dropout (ANN_MCD). An uncertainty model for two important soil properties, creep index C-alpha,C- and hydraulic conductivity k, that control the long-term performance of geotechnical structures is proposed in a function of three soil physical properties using ANN_MCD. Evaluation of the accuracy, uncertainty, and monotonicity of the predicted results for both C-alpha and k reveals the excellent performance of the proposed model, which is used to simulate the long-term settling and excess pore pressure of an embankment on soft clays. The predicted results show good agreement with observations, within a 95% confidence interval. All results indicate that the proposed ANN_MCD-based modelling approach can be used to rapidly correlate soil properties with an uncertainty evaluation and can be further combined with numerical modelling to analyze an engineering-scale problem and conduct risk assessment.

Automated summary

This study introduces a novel approach by utilizing an artificial neural network with Monte Carlo dropout to correlate soil properties with uncertainty. The proposed model shows excellent performance in predicting accuracy, uncertainty, and monotonicity, which can be applied to simulate the long-term settling and excess pore pressure of an embankment on soft clays.