A study on influential rock properties for predicting the longitudinal wave velocity in a rock bolt: Numerical and machine learning approaches

Abstract

The longitudinal wave velocity (vL) in a rock bolt is a useful parameter for evaluating the conditions of the rock bolt and the surrounding rock mass. This study investigated the influence of rock properties on the prediction of vL in a rock bolt using numerical and machine learning approaches. Through numerical simulations, we obtained a dataset of the variations in vL according to rock properties, compressional wave velocity (vp), shear wave velocity (vs), density (ρ), Poisson's ratio (μ), porosity (η), uniaxial compressive strength (UCS), and slake durability index (ISD). This dataset was used to design a deep neural network, and the predicted vL was correlated with rock properties. Notably, vL is strongly correlated with vp, vs, ρ, η, UCS, and ISD. Principal component analysis was employed to characterize the relationship between the rock properties, and the retaining rock properties for random forest (RF) were determined. In the RF, the variable importance (VI) of rock properties was assessed. In particular, vs emerged as the most significant predictor of vL. However, relying on vs to predict vL is not sufficient because it accounts for approximately 60–70 % of the VI. For a more reliable prediction of vL, it is essential to incorporate both vs and vp, which collectively account for approximately 80 % of VI. Notably, the VI of physical properties (vp, vs, ρ, and η) accounts for more than 90 %, implying that these properties can be effectively used to predict vL even in the absence of data concerning mechanical properties.