LAUSR

Abstract The existence of highly complex and heterogeneous geological and hydrogeological conditions makes it cumbersome to determine grouting parameters for a cost-efficient grouting process. Although many empirical, numerical and analytical models have been proposed previously, there are still some gaps between the existing predictive models and practical grouting applications, leading to the fact that practical grouting design mainly depends on onsite engineers’ experience. In this study, we propose to use data mining to discover grouting knowledge from onsite data of a project in Singapore. After systematic analysis of data concerning the geological information, hydrogeological conditions and grouting records, an artificial neural network was structured to further extract grouting knowledge, based on which the grout take can be estimated under given geological and hydrogeological conditions. The grout take at individual station is found to be closely correlated with overall water inflow and Q value of rock mass, making it promising to estimate the potential grout take, once probe hole and face mapping information are given before pre-grouting. The degree of correlation between input parameters and the corresponding model accuracy are significantly affected by the classification methods used.