Assessing Conditioning States of EPB Shield Muck Using Machine Learning–Assisted Calculation Approach for Apparent Density

Soil conditioning plays a vital role as an auxiliary technology for Earth Pressure Balance (EPB) shield tunneling. However, it remains a challenging task to evaluate whether the conditioned soils during tunneling are desired or not. This study proposes a machine learning–assisted apparent density approach to assess the conditioning state of shield muck. Through extensive experimental data, a machine learning–based surrogate model is established to predict the lateral earth pressure coefficient, which enables the calculation of the apparent density of foam-conditioned soils. In addition, the compressibility of foam-conditioned soils is incorporated into the improved evaluation method. To validate this method, the apparent density evaluation is applied to optimize the conditioning scheme for EPB shield tunneling projects in Kunming, China. The results show that the lateral earth pressure coefficients predicted by the random forest algorithm are in good agreement with those unseen compression tests during the training stage. The foam injection ratio is optimized by considering the reasonable range of the apparent density and the suggested minimum compressibility coefficient. After optimization, the advancing speed experiences a striking increase, accompanied by a significant decrease in thrust and torque. The engineering application manifests that the proposed apparent density evaluation method is conducive to improving the excavation efficiency of EPB shield machines.