|
The increasing electrification of aircraft systems and development of autonomous aviation technologies
increase the need for accurate characterization of electromagnetic field (EMF) distributions under operational
conditions. However, dense three-dimensional EMF measurements in aviation environments are difficult to
obtain because airborne observations are typically sparse, irregularly distributed, and collected under varying
flight conditions. Here, we present an artificial intelligence-based framework for three-dimensional reconstruc-
tion of the electric field component of EMF from real-world airborne measurements in general aviation. The
study uses more than 80 h of measurements collected during operational activities involving four aircraft
types and a low-cost, high-sensitivity sensing platform recording synchronized electric-field and geospatial
data over the 1-8000 MHz frequency range. Geostatistical and machine-learning approaches, including Krig-
ing, Kolmogorov-Arnold Networks, and Exact Gaussian Processes, were evaluated for three-dimensional field
reconstruction. Based on this comparison, we developed a hybrid Deep Gaussian Process Kolmogorov-Arnold
Feature Extractor (DGKAN), combining nonlinear feature representation with probabilistic spatial modeling
and uncertainty estimation. DGKAN achieved the highest predictive accuracy, with a mean absolute error of
0.0381 and a mean squared error of 0.0071, reducing the mean absolute error by approximately 27.6% relative
to the strongest baseline model. These results demonstrate that combining Kolmogorov-Arnold representations
with probabilistic Gaussian Process modeling enables accurate reconstruction of three-dimensional electromag-
netic field distributions from sparse airborne measurements. The proposed framework provides a data-driven
approach for electromagnetic environment assessment in general aviation and establishes a basis for further
development of intelligent EMF monitoring and prediction methods under operational flight conditions.
|
