ASH Phase II

Stellar Science previously developed the End-to-End Modeler (ETEM), a high-fidelity laser simulation framework with vital capabilities such as thermal modeling, beam combination, beam propagation (with turbulence and thermal blooming), and adaptive optics. During our Phase I effort for the Advanced Simulation of Hypersonics (ASH), we proposed and began development on adapting ETEM to tackle the difficult problem of intercepting maneuverable hypersonic missiles with directed energy (DE) weapon systems. This development included the addition of Python interfaces, allowing for rapid prototyping of new physical processes that need to be modeled to effectively simulate engagements with hypersonic targets. These interfaces were used to investigate implementation of important processes such as laser propagation through a plasma boundary layer. Our findings will be used to inform prioritization of the most impactful improvements to the ETEM software. Phase I efforts also included the integration of Loci/CHEM into ETEM, aiding with the simulation of laser-plasma interaction. In Phase II, we propose further extension of ETEM’s capabilities through the addition of coarse tracking and gimbal dynamics. These improvements will allow for more accurate assessments of engagements with maneuverable targets. We will also integrate the Aerothermal Targets Analysis Program (ATAP) and the Charring Ablator Response (CHAR) code into our framework. The Python interfaces added in Phase I also allowed for experimentation with machine learning (ML) models designed to quickly and accurately replicate the results of high-fidelity simulations. After identifying several computational bottlenecks that greatly increased the runtimes of ETEM simulations, we architected and prototyped proof-of-concept neural operator-based ML models to accelerate computation of relevant physical phenomena, demonstrating a 50x speedup over traditional solvers. These efforts also resulted in the creation of a generalizable data generation and ML model training pipeline that can be applied to additionally identified future problems such as full atmospheric propagation. In Phase II, we propose expanding upon our framework capabilities to support more diverse model architectures and advanced techniques such as quantization and distributed training. We will also initiate community collaboration on the problem of using ML to aid simulations by hosting an accessible model repository, providing model weights and files as well as details about capabilities, limitations, training, and performance on benchmark metrics. These products will be easily accessible, given GPOC approval, for all Department of Defense (DoD) contractors and agencies under the Joint Directed Energy Transition Office Central initiative. Approved for Public Release | 25-MDA-12220 (23 Sep 25)