Machine Learning-Based Prediction Models of Airblast Overpressure in Tunnel Excavation for Sustainable Development

Blasting is an essential and cost-effective method for rock mass excavation in mining, tunneling, and civil engineering projects. However, it often leads to deleterious environmental issues, particularly airblast overpressure (AOp), which can harm humans, damage structures, and disrupt ecosystems. Accurate AOp prediction is essential for sustainable tunneling practices to mitigate these impacts. This study develops a Levenberg-Marquardt-trained artificial neural network (ANN-LM) model to predict AOp based on 100 blasting events from three tunnel excavation projects in South Korea. Input parameters include charge per delay (Q), rock mass rating (RMR), hole depth (H), tunnel cross-sectional area (TCSA), and distance from blast initiation (DIS). The ANN-LM model was compared with multilinear regression (MLR) and two empirical models. Based on the results, the ANN-LM 5-6-1 demonstrated superior predictive performance with R = 0.8970, RRMSE = 0.08139, and VAF = 81.58. The ANN-LM 5-6-1 model was later transformed into a user-friendly mathematical equation for easier AOp estimation in tunnel excavation. Sensitivity analysis confirmed that RMR, TCSA, DIS, H, and Q are key predictors of AOp. The proposed model offers a reliable, accurate, and efficient method for predicting AOp in tunneling, reducing environmental and structural impacts, and supporting sustainable underground construction.