Dardashti a., A.F.,
Rostami j., ,
Hassanpour j., J.,
Ajalloeian, R.,
Salimi a., Publication Date: 2026
Tunnelling and Underground Space Technology (08867798)172
Given their widespread adoption in numerous rock engineering projects and the availability of influential parameters, various rock mass classification systems serve as suitable methods for estimating Tunnel Boring Machine (TBM) performance. Among these systems, the Rock Mass Rating (RMR) system exhibits a stronger correlation with TBM performance due to its incorporation of uniaxial compressive strength (UCS) as a critical input parameter. This study aims to refine the RMR system to improve TBM performance prediction in hard rock by integrating artificial intelligence algorithms. Originally developed for assessing rock mass stability and designing tunnel support systems, the RMR classification assigns ratings to input parameters based on their influence on stability. However, this inherent focus may explain its limited correlation with TBM performance. To address this limitation, this study proposes an optimized RMR framework by adjusting parameter ratings and their internal weightings to better align with the objective of accurately predicting TBM performance in hard rock conditions. To achieve this, data from ten tunneling projects with diverse geological conditions were compiled into a comprehensive database encompassing geological and geotechnical characteristics of tunnel rock masses, as well as operational and actual machine performance data. The development of the optimized RMR framework involves the application of non-linear regression algorithms, decision trees, and Random Forest models. Ultimately, the refined classification system, RMRTBM, is introduced to enhance the prediction of TBM performance in hard rock. The Random Forest algorithm provides a variable importance index that quantifies the weight assigned to each variable in the model to predict the target variable. This index was used to determine the optimal ratings of RMRTBM input parameters. In RMRTBM, each parameter is assigned a rating based on predefined tables and diagrams, with the total rating ranging from 0 to 100, determining the RMRTBM value. The corresponding RMRTBM class is subsequently used to predict key tunneling performance metrics, including the field penetration index (FPI), boreability classification, TBM excavability description, and stability conditions along the tunnel route. This model offers a valuable tool for improving the design and planning phases of tunneling projects. © 2026 Elsevier Ltd
