Grouting in rock tunnels by data mining 2

Abstract

In rock tunnelling, it is common to encounter water seepage issues, especially for areas with high groundwater tables. Grouting is commonly used by engineers to limit water seepage. The grouting techniques used are predominantly empirical, with some developments in theoretical and analytical approaches. However, these developments are mostly idealistic in nature and are not widely adopted. Therefore, a more efficient method for guidance in grouting works is needed. Conversely, engineering data collected during civil works often remain underutilised.

This research aims to utilise artificial neural network (ANN) models as a data mining approach to uncover grouting knowledge from engineering data collected from the Jurong Rock Caverns project. Specifically, the best way to uncover hidden relationships between hydrogeological parameters and the total grout volume required by tunnel station, in the form of ANN models, is identified by exploring different methods of utilising the collected data as input variables fed into the models. This is supplemented by establishing the most optimal ANN model hyperparameters, namely the number of training epochs and hidden nodes. Furthermore, the influences of each input on the output are analysed for the models, to improve the interpretability and reliability of the results.

Through this research, it is found that ANN models can map and learn the complex relationships between hydrogeological parameters and total grout volume required with reasonable accuracy. Through sensitivity analyses, it is also determined that water inflow rate and water inflow pressure are two of the more significant variables in determining the output. Overall, the research brings about more insights in the guidance of estimating total grout volume for grouting works, using ANNs.