The design and scale-up of fluidized-bed thermochemical reactors are challenging due to the strong coupling between hydrodynamics, heat transfer, and reaction kinetics, as well as the limited transferability of models trained at a single scale. This work proposes a physics-informed CFD–machine learning framework for reactor design and similarity-based scale-up using reduced-order surrogate models. A structured CFD campaign generates a consistent dataset across reactor sizes and operating conditions. Starting from the governing equations, a dimensionless feature space is derived through similarity analysis and used to train Gaussian Process models that predict thermo-fluid reactor descriptors and lumped product yields. Feature relevance is then exploited to support constrained design and to relax full similarity requirements in scale-up applications. Demonstrated on cellulose fast pyrolysis in a fluidized-bed reactor, the methodology improves the interpretability and consistency of surrogate predictions across multiple scales and is formulated to be extendable to other gas-solid thermochemical system.
Baldelli, M., Bartolucci, L., Cordiner, S., Mele, P., Mulone, V. (2026). A similarity-based CFD–machine-learning framework for the design and scale-up of fluidized-bed reactors: application on biomass fast pyrolysis. INTERNATIONAL JOURNAL OF HYDROGEN ENERGY, 255 [10.1016/j.ijhydene.2026.156385].
A similarity-based CFD–machine-learning framework for the design and scale-up of fluidized-bed reactors: application on biomass fast pyrolysis
Baldelli M.
Conceptualization
;Bartolucci L.Writing – Review & Editing
;Cordiner S.Writing – Review & Editing
;Mele P.Validation
;Mulone V.Writing – Review & Editing
2026-07-29
Abstract
The design and scale-up of fluidized-bed thermochemical reactors are challenging due to the strong coupling between hydrodynamics, heat transfer, and reaction kinetics, as well as the limited transferability of models trained at a single scale. This work proposes a physics-informed CFD–machine learning framework for reactor design and similarity-based scale-up using reduced-order surrogate models. A structured CFD campaign generates a consistent dataset across reactor sizes and operating conditions. Starting from the governing equations, a dimensionless feature space is derived through similarity analysis and used to train Gaussian Process models that predict thermo-fluid reactor descriptors and lumped product yields. Feature relevance is then exploited to support constrained design and to relax full similarity requirements in scale-up applications. Demonstrated on cellulose fast pyrolysis in a fluidized-bed reactor, the methodology improves the interpretability and consistency of surrogate predictions across multiple scales and is formulated to be extendable to other gas-solid thermochemical system.| File | Dimensione | Formato | |
|---|---|---|---|
|
1-s2.0-S0360319926030235-main.pdf
accesso aperto
Licenza:
Creative commons
Dimensione
12.98 MB
Formato
Adobe PDF
|
12.98 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.


