Simulation and neural network-based prediction of stress wave attenuation in soil explosions of aluminized charges

The directional differences in the distribution of stress waves generated by explosive charges with varying aspect ratios in soil play a critical role in the design of underground engineering projects as understanding these variations allows for a better prediction of ground motion. To investigate these variations, a method involving borehole charging and sensor installation was developed to monitor both normal and lateral stress dynamics during soil explosions. The transmission characteristics of stress waves generated by TNT and aluminized explosives were recorded. Moreover, a computational model simulating underground explosions was established and calibrated using experimental results, demonstrating a peak deviation of 10.3 %, thereby validating the reliability of the numerical approach, constitutive models, and input parameters. Further simulations revealed the propagation of stress waves generated by aluminized charges with varying aspect ratios, leading to the development of an empirical equation for explosion-induced stress attenuation. Additionally, an artificial neural network (ANN) framework was established to estimate the stress induced by aluminized explosive detonations in soil, incorporating variables such as aspect ratio, azimuthal angle, and scaled distance. Compared with the empirical model, the ANN showed significantly better agreement with the simulation results. To further interpret model predictions, the Shapley Additive Explanations (SHAP) approach was employed to evaluate the contribution of each input variable. Overall, the findings of this study provide a useful foundation for soil explosion studies and the design of subsurface protective infrastructure.