ABSTRACT
The single most significant challenge facing the field of tunneling is not being able to know with a fair degree of certainty what lies underneath the surface. Despite the availability continuous machine performance data recorded by TBMs, and despite much previous research, real time forecasting models for ground conditions still do not exist. Without models that predict geology and the related uncertainties, automation of tunneling construction is nearly impossible, since these models are needed for real time optimization of tunneling operation. In this paper we briefly review the existing research and available predictive Machine Learning models that have been developed to forecast ground during TBM construction, and we list their main limitations. Suggestions for future research, are provided to develop more robust and generalizable geological forecasting models. Insights from a case study are used to illustrate two potential solutions to improve ML model prediction performance.
