Toward machine learning based decision support for pre-grouting in hard rock

Abstract

Pre-grouting in hard rock tunneling is crucial for mitigating water ingress, significantly affecting project time and cost. Predicting pre-grouting requirements is challenging and relies heavily on the expertise of on-site personnel for decision-making. This paper explores using supervised machine learning (ML) to create a data-driven pre-grouting decision process, aiming to predict “grouting time” and “total grout take.” Tree-based regression models were developed using data from a Norwegian railway project, including typical tunneling data. These models showed limited predictive performance, with R² scores of 0.40, though a significant relationship was observed. The limited performance highlights the need to identify parameters that significantly impact grouting outcomes rather than indicating the unsuitability of tree-based models. Future research should consider a larger data set and additional parameters, such as more data on rock mass quality, hydrogeological conditions ahead of the face, and human, organizational, and contractual factors. Despite initial findings, supervised ML shows promise in enhancing data-driven decision-making in pre-grouting by using appropriate input features and target variables.