Chisel bits cutting force estimation using XGBoost and different optimization algorithms

Mohammad Rouhani

doi:10.1016/j.compgeo.2024.106465

2024-08-01

PRE

Mohammad Rouhani

Ebrahim Farrokh *

DOI：10.1016/j.compgeo.2024.106465

原文链接

原文求助

摘要

En 中文

The design and selection of tools for mechanized excavations is a crucial topic in the field of tunneling. Chisel bits are commonly used tools for soft ground, with the required cutting force being a vital factor in tool design. However, current methods either lack accuracy or fail to consider all necessary parameters. To address this issue, a new cutting force prediction model was developed to improve response quality and consider the most effective parameters. 206 data points were collected from studies on rock cutting with different strength, including information such as Uniaxial compressive strength, Brazilian tensile strength, Depth, Width, Rake angle, and clearance angle. Five different algorithms were then used to optimize the Hyperparameters in the XGBoost method, including Grid Search, Random Search, Bayesian Algorithm, Differential Evolution, and Optuna. Results showed that while most algorithms provided appropriate responses, the Grid Search and Bayesian Algorithm methods were the most effective, with R 2 values (in test and train) of 0.876 and 0.93 in GS and 0.872 and 0.926 in BO, respectively. Upon comparison of the two methods, it was discovered that the GS approach yields a superior solution; however, it is also more sensitive to tuning, requires more calculations, and takes longer to provide a solution. When assessing the newly presented approach against existing methods, it was noted that the Evans and Roxborough methods produced R 2 values of 0.48 and 0.53, respectively, which are notably lower than those of the new method. Lastly, the parametric analysis revealed that the cutting force is primarily affected by the depth, width, and rake angle in that order.

Keyword:

Cutting force

Chisel bit

XGBoost

Grid search

Bayesian algorithm