Applicability of existing criteria of rockburst tendency of sandstone in coal mines

Abstract

To evaluate the accuracy of rockburst tendency classification in coal-bearing sandstone strata, this study conducted uniaxial compression loading and unloading tests on sandstone samples with four distinct grain sizes. The tests involved loading the samples to 60%, 70%, and 80% of their uniaxial compressive strength, followed by unloading and reloading until failure. Key parameters such as the elastic energy index and linear elasticity criteria were derived from these tests. Additionally, rock fragments were collected to calculate their initial ejection kinetic energy, serving as a measure of rockburst tendency. The classification of rockburst tendency was conducted using grading methods based on burst energy index (W_ET), pre-peak stored elastic energy (PES) and experimental observations. Multi-class classification and regression analyses were applied to machine learning models using experimental data to predict rockburst tendency levels. A comparative analysis of models from two libraries revealed that the Random Forest model achieved the highest accuracy in classification, while the AdaBoost Regressor model excelled in regression predictions. This study highlights that on a laboratory scale, integrating ejection kinetic energy with the unloading ratio, failure load, W_ET and PES through machine learning offers a highly accurate and reliable approach for determining rockburst tendency levels.