Abstract
Determining the rock mass rating (RMR) plays a key role in designing rock engineering projects. Direct determination of the RMR involves the measurement of six parameters which is a destructive, expensive and time-consuming task. Hence, non-destructive determination of the RMR based on the easy and inexpensive measurable parameters can be an attractive and economic process. In this study, relationships of the RMR with shear wave velocity (V(s)) and compression wave velocity (V(p)) of the rock mass are evaluated using genetic algorithm (GA), trust region reflective (TRR), and hybrid TRR-GA model. This study involves the analysis of 150 in-situ datasets related to the rock mass with diverse rock types. Random forest (RF) analysis confirmed that the simultaneous combination of both V(p) and V(s) parameters is more reliable for RMR determination than when each of V(p) and V(s) is considered individually. Deep estimation capability analyses of the proposed GA, TRR and GA-TRR models were performed using the performance evaluation metrics, scatter plots, error histogram, Taylor diagram and regression error characteristic curve. Results indicated that all suggested models provide high accuracy in predicting RMR. However, the hybrid TRR-GA model emerging as the best model in predicting RMR based on the V(p) and V(s). The diversity of examined rock types, utilizing the cost-effective and easily measurable input variables, and the application of non-destructive robust meta-heuristic algorithms for RMR determination are the main innovations of this study. However, further studies involving more datasets and diverse rock types are required for more validation and practical application of these findings.