Machine learning meso scale structure property performance relationships of energetic materials for multiscale modeling of shock induced detonation

Abstract

Performance of energetic materials such as propellants, explosives and pyrotechnics depends on local energy release phenomena at the meso scale, i.e. the scale of the particles and defects in the materials.However, to a designer, the overall system scale response of the energetic material is of interest. To predict the behavior of process scale systems that rely on energetic material response to loads, the smallscale physics must be correctly reflected in the large scale dynamics calculations. This calls for a multiscale approach connecting the small (meso) scales to the large (macro or process) scales. The proposed project seeks to advance the state of the art of multiscale modeling of heterogeneous energetic materials (HEs) by using machine learning techniques to represent and generalize the behavior of a wide class of commonly used nitramine based HEs. This proposal will advance modeling by extending a well developed capability for simulation of imaged 3 dimensional meso structures by using a state of the art computational code called SCIMITAR3D. Machine learning techniques, such as Kriging based surrogate modeling and Deep Neural Networks will be employed to learn the behavior at the meso scale and to construct functional relationships between mesostructure, material properties and performance of the HEs. It will also advance reliability by providing tools to bound uncertianties in the predictions by relating input parameter uncertainties in the inherently stochastic system to uncertainties in predicted performance. The combination of accurate computation of the shock response of heterogenous materials, surrogate model construction to link scales, uncertainty quantification in the multi scale framework and machine learning to assimilate behavior of a class of HEs will lead to a comprehensive predictive capability for HE performance. We expect that the tool developed as a result of the proposed work will be an integral part of an explosive and munitions design.

Document Details

Document Type
DoD Grant Award
Publication Date
Jan 14, 2022
Source ID
FA95501910238

Entities

People

  • H. S. Udaykumar

Organizations

  • Air Force Office of Scientific Research
  • United States Air Force
  • University of Iowa

Tags

Readers

  • Computational Fluid Dynamics (CFD)
  • Neural Network Machine Learning.
  • Rocket Propulsion.

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference
  • AI & ML - Neural Networks