Estimation of foam (surfactant) consumption in earth pressure balance tunnel boring machine using statistical and soft-computing methods

Abstract

The use of foam, as the most economical soil conditioning technique, in earth pressure balance tunnel boring machine (EPB-TBM) tunneling projects has significant effects on operation efficiency, excavation cost, and operation time. This study mainly focuses on developing models to predict the foam (surfactant) consumption. For this purpose, five empirical models are developed based on a database containing 11048 datasets of real-time foam consumption from three EPB-TBM tunneling projects in Iran. This database includes the most effective machine operational parameters and soil geomechanical properties on the foam consumption. Multiple linear regression analysis, multiple non-linear regression analysis, M5Prime decision tree, artificial neural network, and least squares support vector machine techniques are used to construct the models. To evaluate the performance of developed models, three performance evaluation criteria (including normalized root mean square error, variance account for, and coefficient of determination) are used based on the training and testing datasets. The results show that the developed models have high performance and their validity is guaranteed according to the testing dataset. Furthermore, the M5Prime model, which demonstrates the best performance compared to other models, is applied to predict the foam consumption in 19 excavation rings of Kohandezh station in Isfahan metro, Iran. After conducting an excavation operation in this station and comparing the results, it was found that the M5Prime model accurately predicts foam consumption with an average error of less than 13%. Therefore, the developed models, particularly M5Prime model, can be confidently applied in EPB-TBM tunneling projects for predicting foam consumption with a low error rate.