Artificial Neural Network (ANN) model for predicting blast-induced tunnel response in Steel Fiber Reinforced Concrete (SFRC) structures

Abstract

This study presents an Artificial Neural Network (ANN)-based predictive framework for evaluating the blast-induced response of Steel Fiber Reinforced Concrete (SFRC) tunnel structures. As underground infrastructure is increasingly exposed to dynamic and extreme loading conditions, particularly from accidental or intentional explosions, accurate and efficient prediction tools are essential. In this research, a comprehensive dataset comprising 299 data points was developed, including approximately 120 experimental results from published blast and structural tests, and 179 high-fidelity numerical simulations. This combined dataset ensured both physical reliability and broad coverage of loading scenarios. The model incorporates nine critical input parameters: Peak Overpressure (MPa), Impulse (kPa & sdot;ms), Tunnel Diameter (m), Wall Thickness (m), Compressive Strength (MPa), Tensile Strength (MPa), Fiber Volume Fraction (%), Soil Stiffness (MPa/m), and Standoff Distance (m). The target output variable is the tunnel's Maximum Displacement (mm) under blast loading. A three-hidden-layer ANN architecture was optimized through rigorous hyperparameter tuning. The best-performing model, with 16 neurons in each hidden layer, achieved high predictive accuracy, with R2 values of 0.983 (training), 0.956 (validation), and 0.948 (testing). Error metrics including RMSE (2.12-3.14 mm), MAE (1.92-3.52 mm), and MAPE (1.95 %-3.12 %) further confirmed the model's robustness. Validation against experimental data from literature demonstrated excellent agreement, verifying the model's practical applicability. Additionally, sensitivity analysis identified Peak Overpressure and Standoff Distance as the most influential factors affecting displacement. The proposed ANN framework offers a computationally efficient and accurate tool for assessing SFRC tunnel performance under blast loading, supporting the design of safer and more resilient underground structures.