Learning robust parameter inference and density reconstruction in flyer plate impact experiments.

Area of Science:

• Physics
• Material Science
• Computational Science

Background:

• Radiography is crucial in shock physics experiments but doesn't directly yield key variables like density.
• Traditional parameter estimation fails when direct access to state variables is limited.

Purpose of the Study:

• To develop a machine learning (ML) approach for inferring material properties from radiographic data.
• To address limitations in estimating equation of state (EoS) and crush porosity parameters from radiography.

Main Methods:

• Utilized flyer plate impact experiments on porous materials.
• Employed generative machine learning to produce posterior distributions of physical parameters directly from radiographs.
• Combined low and high impact velocity data to capture different compaction and shock propagation regimes.

Main Results:

• Demonstrated that high-velocity data alone is insufficient for accurate parameter inference.
• Showcased the ML approach's effectiveness in estimating EoS and crush model parameters from simulated experiments.
• Validated that estimated parameters improve density reconstructions in hydrodynamic simulations.

Conclusions:

• The proposed ML approach enables accurate material property estimation from radiographic observations.
• The method is robust to noise and model mismatches, offering a potential breakthrough in experimental data analysis.
• This technique facilitates improved understanding and prediction of material behavior under extreme conditions.