The Confluence of Science- and Machine-Based Learning Approaches in Energetic Materials Research

Leading subject matter experts in Energetic Material (EM), Reactive Material (RM), and Machine-based Learning (ML) disciplines will come together to discuss recent discoveries and cutting-edge research. If one visualizes a Venn diagram, the overlap area of these three R&D subject areas offers opportunity to accelerate a deeper understanding of EM/RM physics, chemistry, and kinetics. Moreover, machine-based results motivate human-based R&D.

Large-scale EM/RM empirical performance tests verify and validate (V&V) modeling predictions and can subsequently illuminate fundamental knowledge gaps. However, such test campaigns are challenging, dangerous, and costly. Furthermore, they may not bracket conditions/materials of future interest. Moreover, the task of uniquely attributing V&V results to fundamental chemical physics phenomena, at nano and meso spatial scales, is frequently fraught; enduring questions abound concerning uniqueness in attempts to relate macro-scale observations to the underlying physical and chemical mechanisms (i.e., the uniqueness of solutions to the inverse problem).

Semi-empirical physics- and chemistry-based modeling and simulation (M&S) effort, anchored to sub-scale experimental data, provide a more commonly accessible and robust basis for confident explosives safety and performance predictions. While this is a proven reliable path forward, technology advancement could be accelerated.

Rapidly emerging advances in ML approaches hold promise for substantially alleviating the decades-long impediment to robust and reliable science-based predictive capability. ML publications seemingly account miraculous optimization routes identifying, characterizing, and quantifying couplings among chemistry, meso-scale chemical physics and engineering mechanics, and experimentally measured observation. Energetic materials and RM research and engineering are on the cusp of disruptive (transformational) change initiated by data-assimilation in M&S codes including V&V data generation, and the predictive capabilities of neural networks, genetic algorithms, deep learning, and additional ML approaches. Our 2022 EM GRC and GRS are structured with traditional topic areas with presentations formally acknowledging we are at a new confluence of research disciplines. Come join us and learn, teach, and develop fruitful collaborations!