Investigation on Physics Aware Neural Networks for the Determination of Detonation Parameters of Non-Ideal Explosives

The development and optimization of weapon systems containing explosives have necessitated the utilization of numerical simulations as an indispensable instrument. Beyond numerical precision, modeling error, particularly that pertaining to input data, has emerged as a critical component in evaluating the accuracy of simulations. For non-ideal explosives, the generation of input data can be facilitated through the application of Wood and Kirkwood's theory. However, this approach can impose substantial computational demands due to its reliance on a system of hyperbolic nonlinear partial differential equations (PDEs), which can be computationally intensive. Physics-aware neural networks present a promising avenue for leveraging highly optimized algorithms on existing hardware architectures, thereby serving as a potential alternative to traditional numerical algorithms as fast, standalone solvers. The objective of this research is to investigate the viability of a thermodynamics code for the calculation of detonation parameters of non-ideal explosives using GPU-accelerated, physics-aware neural networks.