Physics-embedded probabilistic model for extrapolating the electromagnetic scattering from a metal target under...

Area of Science:

• Electromagnetics
• Computational Physics
• Machine Learning

Background:

• Radar cross section (RCS) extrapolation is crucial for target analysis.
• Existing RCS extrapolation models often lack fidelity and reliability.
• Modeling electromagnetic wave interactions requires physics-informed approaches.

Purpose of the Study:

• To develop a physics-embedded Gaussian process regression (GPR) model for high-fidelity and reliability-aware RCS extrapolation.
• To establish a novel GPR framework in the scattered electric-field domain, leveraging electromagnetic wave properties.
• To integrate physical priors into GPR for improved extrapolation accuracy.

Main Methods:

• A composite kernel integrating a degree-2 generalized polynomial for scaling laws and a spectral mixture (SM) kernel for interference effects was designed.
• The GPR model was established in the scattered electric-field domain, not the RCS domain.
• The model was validated on a warhead model and the SLICY model.

Main Results:

• The proposed physics-embedded GPR significantly reduces root-mean-square extrapolation error by up to 91.5% compared to alternative models.
• The model demonstrates high fidelity in extrapolating RCS under geometric scaling.
• Physics-grounded uncertainty quantification was provided for reliability assessment.

Conclusions:

• Physics-embedded GPR offers a superior method for RCS extrapolation, enhancing both accuracy and reliability.
• Operating in the scattered electric-field domain provides a more physically grounded approach.
• The integrated composite kernel effectively encodes physical laws and interference effects.