Abstract The present study examines the mechanical properties of intact sandstone and shale from the Banhardi Coal Block (48–819 m depth) under both dry and saturated conditions, which are essential for deep mining applications. Laboratory analyses quantified uniaxial compressive strength (UCS), elastic modulus, density, porosity, and water absorption, while statistical and machine learning methodologies examined the interrelationships among these properties. Results indicate that water saturation consistently diminishes rock strength and stiffness. Depth-property analysis indicated: (1) UCS exhibits linear trends in shale (R2 = 0.327) and power–law behaviour in sandstone (R2 = 0.3698); (2) Elastic modulus demonstrates a more pronounced depth-dependence in saturated sandstone (R2 = 0.4161); (3) Porosity consistently diminishes with depth (R2 = 0.42–0.48). Although multivariate regression revealed significant connections, nonlinearities constrained its predictive efficacy. A comparative investigation indicated that genetic algorithm–optimized artificial neural network (GA-ANN) models (R > 0.85) surpassed both KNN and regression methods, exhibiting enhanced accuracy in representing saturation effects and depth-dependent behaviour with minimum error (low RMSE/WMAPE). The research identifies GA-ANN as an effective instrument for geomechanical forecasting in heterogeneous layers, especially for evaluating water-weakening impacts in deep formations. These findings enhance predictive modelling for subterranean construction, emphasising the joint impact of depth and saturation on rock characteristics. Research highlights <UnorderedList Mark="Bullet"> <ItemContent> Intact sandstone and shale from 48 to 819 m depth were tested in dry and saturated states to quantify coupled effects of depth, porosity, and saturation on UCS and modulus of elasticity. </ItemContent> <ItemContent> Multivariate regression captured only part of the variability (R2 up to ~0.62), whereas GA-optimized ANN models achieved overall correlation coefficients above 0.85 for all cases, significantly improving strength and stiffness predictions. </ItemContent> <ItemContent> Depth shows power–law or logarithmic relations with UCS and E, while porosity systematically decreases with depth; saturation causes consistent strength and stiffness degradation across both lithologies. </ItemContent> <ItemContent> GA-ANN outperforms KNN and classical regression in terms of R, RMSE, WMAPE, and VAF, making it a robust tool for predicting rock mechanical properties in heterogeneous, water-affected deep formations. </ItemContent> </UnorderedList>

Predicting intact rock strength for mechanical excavation in dry and saturated condition using multivariate statistics and artificial neural networks optimized using genetic algorithm

期刊论文
发表日期：

2026年6月1日

查看全文

Sufi Md Gulzar,
L B Roy,
Sivaji Lahiri,
Pappu Rabidas

摘要

Abstract

The present study examines the mechanical properties of intact sandstone and shale from the Banhardi Coal Block (48–819 m depth) under both dry and saturated conditions, which are essential for deep mining applications. Laboratory analyses quantified uniaxial compressive strength (UCS), elastic modulus, density, porosity, and water absorption, while statistical and machine learning methodologies examined the interrelationships among these properties. Results indicate that water saturation consistently diminishes rock strength and stiffness. Depth-property analysis indicated: (1) UCS exhibits linear trends in shale (R² = 0.327) and power–law behaviour in sandstone (R² = 0.3698); (2) Elastic modulus demonstrates a more pronounced depth-dependence in saturated sandstone (R² = 0.4161); (3) Porosity consistently diminishes with depth (R² = 0.42–0.48). Although multivariate regression revealed significant connections, nonlinearities constrained its predictive efficacy. A comparative investigation indicated that genetic algorithm–optimized artificial neural network (GA-ANN) models (R > 0.85) surpassed both KNN and regression methods, exhibiting enhanced accuracy in representing saturation effects and depth-dependent behaviour with minimum error (low RMSE/WMAPE). The research identifies GA-ANN as an effective instrument for geomechanical forecasting in heterogeneous layers, especially for evaluating water-weakening impacts in deep formations. These findings enhance predictive modelling for subterranean construction, emphasising the joint impact of depth and saturation on rock characteristics.

Research highlights

Intact sandstone and shale from 48 to 819 m depth were tested in dry and saturated states to quantify coupled effects of depth, porosity, and saturation on UCS and modulus of elasticity.

Multivariate regression captured only part of the variability (R² up to ~0.62), whereas GA-optimized ANN models achieved overall correlation coefficients above 0.85 for all cases, significantly improving strength and stiffness predictions.

Depth shows power–law or logarithmic relations with UCS and E, while porosity systematically decreases with depth; saturation causes consistent strength and stiffness degradation across both lithologies.

GA-ANN outperforms KNN and classical regression in terms of R, RMSE, WMAPE, and VAF, making it a robust tool for predicting rock mechanical properties in heterogeneous, water-affected deep formations.