IGNEOUS

Surrogate ML Model for Magneto‑Hydrodynamics Prediction

Built an FNN surrogate to emulate MHD simulations with >98% fidelity, reducing turnaround to <20 minutes.

January 1, 2025
Neural Network MATLAB Python Plasma Physics Machine Learning Computational Physics Simulation

Problem

Accelerate computationally expensive MHD simulations for rapid design iteration.

Radial Implosion trajectory prediction from current pulses

Approach

  • Collected simulation inputs/outputs and standardized features for learning.
  • Trained a feed‑forward network surrogate to regress key MHD fields/metrics.
  • Validated fidelity against reference runs and profiled end‑to‑end runtime.

Inverse Problem - Predicting Current Pulses from Radial Trajectories Inverse Problem - Predicting Current Pulses from Radial Trajectories

Impact

  • Delivered >98% agreement on target metrics with orders‑of‑magnitude faster runtime.
  • Enabled rapid HED design studies and parameter sweeps.

Tech

  • Modeling: FNN regressors, regularization, early stopping
  • Tooling: MATLAB prototyping, Python packaging