Abstract:
Reinforcement learning (RL), a branch of machine learning, incorporating human-level
control, is attracting considerable interests in many fields. Currently, there is a transition
from academia to real-world prototypes, with RL-examples like the optimization of a
manufacturing process, for real-time steering of hydrocarbon drilling, in optimization of
power grids and control systems in general, for UAVs, and in robotics and other
autonomous vehicles becoming common. However, there is very little published
application of RL to geotechnics in general. Therefore, in this study will present a novel
RL-based framework for construction process optimization whereby displacements would
be optimized by application of Hoeke-Brown criterion and GSI system to excavation
design. Such models can act as decision support for the geotechnical engineer, engineering
geologist, geotechnician etc. (design choices, progress-planning) and in the long run such
models work towards full automation in underground construction. Hence, the model is
one of the first attempt to automate decisions made by the geotechnician in excavation
design for underground construction.
