Estimating Pillar Strength for Rock Salt Mines of the Salt Range Pakistan Using Statistical and Artificial Neural Network Modeling Techniques

This research proposes empirical models to estimate pillar strength by adopting multilinear regression and artificial neural network approaches for rock salt mines of the Salt Range, Punjab, Pakistan. The field data of a total of 168 pillars was collected from three (03) selected rock salt mines being operated by Pakistan Mineral Development Corporation. The field work included geometry of pillars, Schmidt rebound hardness (SRH), uniaxial compressive strength (UCS), fracture spacing, fracture condition, joint-orientation, groundwater state, weathering effects, blasting effects, and mining-induced stress. The dataset collected from the field for each rock salt pillar was further utilized to determine rock quality designation (RQD), rock mass rating (RMR), mining rock mass rating (MRMR), design rock mass strength (DRMS), and pillar strength (σ_p). The modeling was done using a dataset of 150 columns, and the remaining data of 18 pillars was left for validation purposes. The proposed ANN and MLR models have R-square (R²) values of 95.35% and 91.61%, respectively. Further, the prediction performance of the ANN model was also compared with that of multilinear regression (MLR). It was found that the ANN model outperformed the MLR model.