Physics consistent machine learning framework for inverse modeling with applications to ICF capsule implosions.

Area of Science:

• High Energy Density Physics (HEDP)
• Inertial Confinement Fusion (ICF)
• Computational Physics

Background:

• Predictive modeling in HEDP and ICF is hindered by unobservable parameters like material properties and equation of state (EOS).
• Radiographic projections are the primary observable data, offering indirect insights into system parameters.

Purpose of the Study:

• To develop a machine learning (ML) framework for inferring unobservable parameters directly from radiographic measurements.
• To reconstruct thermodynamically and hydrodynamically consistent density fields from noisy radiographic data.

Main Methods:

• A two-stage ML pipeline: a radiograph-to-features network (R2FNet) and a features-to-parameters network (F2PNet).
• Extraction of sparse hydrodynamic features from shock profiles and material edges in radiographs.
• Training ML models to approximate posterior distributions for parameters from radiographs.

Main Results:

• Accurate inference of initial conditions and EOS parameters using the ML framework.
• Successful reconstruction of density fields, shocks, and material interfaces consistent with physical laws.
• Demonstrated invariance to underlying EOS models, indicating learning of physical principles.

Conclusions:

• The developed ML framework enables direct parameter inference from radiographs in HEDP and ICF.
• This approach achieves the first demonstration of recovering consistent density fields from noisy radiographs.
• The methodology holds potential for improving the accuracy and reliability of predictive modeling in fusion research.