Three-Dimensional Reconstruction of Fluidized Beds from Limited Measurements via a Physics Guided cGAN

Characterizing multiphase flows in fluidized beds remains challenging due to complex hydrodynamics, harsh operating conditions, and sensor limitations that restrict data acquisition. A physics guided conditional Generative Adversarial Network (PG-cGAN) framework is designed to reconstruct detailed three-dimensional (3D) fields from limited measurements. This hybrid approach integrates a gap-filling preprocess to mitigate data incompleteness, employs a multiscale strategy to enhance feature extraction from sparse inputs, and incorporates empirical correlations as physical constraints to ensure realistic reconstructions. Validation against computational fluid dynamics (CFD) simulations demonstrates close agreement between reconstructed and ground-truth fields. Furthermore, application of the PG-cGAN to 2D cross-sectional slices obtained from mobile Electrical Capacitance Tomography (ECT) experiments enables 3D fluidization analysis. The PG-cGAN framework provides potential for online characterization of complex flows by enabling rapid reconstruction from limited data.