MRI-validated CFD–DEM simulation and recurrence-based time extrapolation (rCFD) of a bubbling fluidized bed: Drag model selection and computational speed-up
Citation Link: https://doi.org/10.15480/882.16945
Publikationstyp
Journal Article
Date Issued
2026-02-19
Sprache
English
Author(s)
TORE-DOI
Volume
227
Start Page
879
End Page
893
Citation
Chemical Engineering Research and Design 227: 879–893 (2026)
Publisher DOI
Scopus ID
Publisher
Elsevier
CFD–DEM simulations of industrial fluidized beds suffer from a major computational bottleneck: simulating one second of physical time often requires hours of wall-clock time on modern computer hardware, with hundreds of processors, which makes long-term process analysis impractical. Recurrence-based CFD (rCFD) addresses this challenge by exploiting pseudo-periodic flow patterns to extrapolate short CFD–DEM simulations over extended time periods, but experimental validation of such time-extrapolated predictions has remained absent. To our knowledge, this is the first rigorous experimental validation of rCFD against internal full-field magnetic resonance imaging (MRI) measurement of a cylindrical fluidized bed (96 mm inner diameter, 600 mm
height) operated with poppy seeds (𝑑32 = 1.16 mm) at 1.5 𝑢𝑚𝑓 . First, CFD–DEM predictions were validated against MRI measurements by comparing four drag models (Koch & Hill, Di Felice, Gidaspow, Beetstra) across multiple metrics: bed expansion dynamics, bubble size distributions, radial particle distributions, and bubble rise velocities. The Koch & Hill model demonstrated superior agreement, capturing mean bed heights within a deviation of 0.9 cm of MRI data (17.7 cm vs. 16.8 cm respectively) and correctly predicting most of the bubble formation patterns observed experimentally. Fast Fourier Transform analysis of the validated CFD–DEM data revealed a characteristic recurrence period of 0.27 s, enabling construction of a 5 s rCFD database. Time-extrapolated rCFD simulations to 10 s maintained excellent agreement with extended MRI measurements, preserving bed expansion behavior, bubble size–velocity correlations without drift or spurious behavior. The approach achieved a 4, 050× computational speed-up, reducing the wall-clock time from 4.5 h to 4 s per simulated physical second while pursuing predictive accuracy. This validated methodology enables previously intractable applications, including real-time process optimization, parametric design studies, and digital twin development for industrial fluidized bed setups.
height) operated with poppy seeds (𝑑32 = 1.16 mm) at 1.5 𝑢𝑚𝑓 . First, CFD–DEM predictions were validated against MRI measurements by comparing four drag models (Koch & Hill, Di Felice, Gidaspow, Beetstra) across multiple metrics: bed expansion dynamics, bubble size distributions, radial particle distributions, and bubble rise velocities. The Koch & Hill model demonstrated superior agreement, capturing mean bed heights within a deviation of 0.9 cm of MRI data (17.7 cm vs. 16.8 cm respectively) and correctly predicting most of the bubble formation patterns observed experimentally. Fast Fourier Transform analysis of the validated CFD–DEM data revealed a characteristic recurrence period of 0.27 s, enabling construction of a 5 s rCFD database. Time-extrapolated rCFD simulations to 10 s maintained excellent agreement with extended MRI measurements, preserving bed expansion behavior, bubble size–velocity correlations without drift or spurious behavior. The approach achieved a 4, 050× computational speed-up, reducing the wall-clock time from 4.5 h to 4 s per simulated physical second while pursuing predictive accuracy. This validated methodology enables previously intractable applications, including real-time process optimization, parametric design studies, and digital twin development for industrial fluidized bed setups.
DDC Class
660.284: Chemical Reactors
Funding(s)
Publication version
publishedVersion