Work

Prediction of soil conditioner dosages for shield tunneling in sandy soil based on machine learning

Year: 2025

Type: article

Abstract: During the construction in sandy soil stratum using Earth Pressure Balance Shield (EPBS) machines, issues such as screw conveyor spewing and cutter wear frequently occur due to inadequate conditioning of the excavated soil. This study develops a data-driven model for optimizing soil conditioner dosages during EPBS tunneling in sandy strata of Shenyang Metro Line 4. Integrating 15 key parameters (shield, geological, geometric), the model analyzed 5,496 ring segments categorized. Spearman analysis revealed strong shield tunneling correlations and extremely strong geological correlations, with PCA extracting six principal components as model inputs. The predictive models (SVR, XGBoost, LightGBM) constructed using excellent datasets and optimized via PSO, BO, and Optuna hyperparameter tuning algorithms demonstrated that the LightGBM-Optuna ensemble model exhibited superior comprehensive performance, achieving 45-58% error reduction relative to baseline configurations. Application to suboptimal datasets decreased specific energy consumption by 13.15% through torque optimization, validating model validity. The model enables intelligent dosage decisions for improved advance rates and reduced cutter wear in complex strata.

Source: Case Studies in Construction Materials

DOI: 10.1016/j.cscm.2025.e05660

Authors: Keqi Liu, Xueyi Liang, Wen Zhao, Xiumei Cao, Hezheng Pei

Institution: Northeastern University

Language: English

Topic: Tunneling and Rock Mechanics

Subfield: Civil and Structural Engineering

Field: Engineering

Domain: Physical Sciences

SDG: Industry, innovation and infrastructure

Open Access status: gold

Estimated APC paid: $2,660