Machine Learning-Based Uniaxial Compressive Strength Estimation for Lignite in an Underground Coal Mine

Abstract

Uniaxial compressive strength (UCS) is one of the most fundamental parameters used in rock mechanics and mining design; however, laboratory UCS testing is often time-consuming, costly, and impractical for continuous field applications. This study aims to develop a rapid and low-cost UCS estimation framework using two easily obtainable indices: Schmidt hammer rebound hardness (SHT) and point load strength (PLT). A total of 114 coal samples collected from the A1 panel of the & Ouml;merler Mine were used to train and evaluate four machine-learning models; multiple linear regression (MLR), regression trees (RT), support vector regression (SVR with linear, polynomial, and RBF kernels), and artificial neural networks (ANN). Model performances were assessed through 5-fold cross-validation and statistically compared using the Friedman and Nemenyi tests. The ANN model achieved the highest predictive accuracy, with an R-2 value exceeding 0.85 and the lowest error metrics among all evaluated algorithms. SVR models also produced competitive results. Statistical rank comparisons confirmed the significant superiority of the ANN model over the RT method. The findings demonstrate that reliable UCS prediction can be achieved using only SHT and PLT, offering a practical and cost-effective alternative for preliminary geotechnical characterization in mining operations. The proposed framework provides field engineers with a fast decision-support tool for strength estimation when laboratory testing is limited or unavailable.