Bayesian deep learning for uncertainty quantification and prediction of jet grout column diameter

Deterministic machine learning methods have been widely utilized for predicting the diameter of jet grout columns. However, these methods often do not account for uncertainties associated with predictions, which are notably variable in practical engineering. To address this limitation, this study presents a framework that utilizes a Bayesian deep neural network to predict the diameter of a jet grout column and quantify the associated uncertainties. A dataset comprising 160 field-measured jet grouting cases was compiled from published case histories. These case histories include details on the ground conditions obtained by Standard Penetration Test (SPT), specific energy utilized during jet grouting, and the resulting jet grout column diameters. The Bayesian deep neural network models were trained, tested, and validated using this dataset, along with uncertainty quantification considering both aleatoric and epistemic sources. The developed model predictions of jet grout column diameters were compared with the diameter predicted using the empirical equation. Design charts were prepared to support optimal jet grouting design, enhancing the accuracy and reliability of jet grout column diameter predictions.