Pyrolysis of residual lignocellulosic biomass plays a crucial role in mitigating climate change by providing a sustainable pathway for carbon-neutral energy production. In particular, fast pyrolysis on luidized-bed reactors is a promising route to renewable bio-oil and syngas, yet reactor design is complicated by the multiphase hydrodynamics and tight temperature-time windows needed to maximize liquid yields. A second key aspect for pyrolysis reactors is the lexibility. The performance of such systems is highly dependent on the composition of the feedstock, which strongly inluences both the distribution and quality of the resulting products. Conventional design techniques rely on both empirical correlations and CFD models, which are computationally expensive and sensitive to uncertain feedstock properties. Recent advances in Machine Learning provide a data-driven alternative that can map complex, nonlinear relationships between operating variables and performance, with a time-scale of several orders of magnitude smaller than standard methods, enabling fast design procedures and reactor real-time control. In this work, the data from an extensive computational luid dynamics (CFD) campaign on pyrolysis of lignocellulosic biomass in luidized-bed reactors are used to train two Neural Networks (NNs): one aimed at predicting the composition of pyrolysis products (suitable for control application), and the other at providing reactor design guidelines based on the desired product output. The NNs test results indicate that the average prediction error is below 5 % relative to the actual target value, demonstrating the robustness of the models over the entire data-set.
Baldelli, M., Bartolucci, L., Cordiner, S., Mulone, V. (2025). Pyrolysis of biomass in fluidized-bed reactors: Towards a data-driven approach for reactor design and control. JOURNAL OF PHYSICS. CONFERENCE SERIES, 3143(1) [10.1088/1742-6596/3143/1/012010].
Pyrolysis of biomass in fluidized-bed reactors: Towards a data-driven approach for reactor design and control
Baldelli M.
Conceptualization
;Bartolucci L.Writing – Review & Editing
;Cordiner S.Supervision
;Mulone V.Writing – Review & Editing
2025-12-15
Abstract
Pyrolysis of residual lignocellulosic biomass plays a crucial role in mitigating climate change by providing a sustainable pathway for carbon-neutral energy production. In particular, fast pyrolysis on luidized-bed reactors is a promising route to renewable bio-oil and syngas, yet reactor design is complicated by the multiphase hydrodynamics and tight temperature-time windows needed to maximize liquid yields. A second key aspect for pyrolysis reactors is the lexibility. The performance of such systems is highly dependent on the composition of the feedstock, which strongly inluences both the distribution and quality of the resulting products. Conventional design techniques rely on both empirical correlations and CFD models, which are computationally expensive and sensitive to uncertain feedstock properties. Recent advances in Machine Learning provide a data-driven alternative that can map complex, nonlinear relationships between operating variables and performance, with a time-scale of several orders of magnitude smaller than standard methods, enabling fast design procedures and reactor real-time control. In this work, the data from an extensive computational luid dynamics (CFD) campaign on pyrolysis of lignocellulosic biomass in luidized-bed reactors are used to train two Neural Networks (NNs): one aimed at predicting the composition of pyrolysis products (suitable for control application), and the other at providing reactor design guidelines based on the desired product output. The NNs test results indicate that the average prediction error is below 5 % relative to the actual target value, demonstrating the robustness of the models over the entire data-set.| File | Dimensione | Formato | |
|---|---|---|---|
|
Baldelli_2025_J._Phys.:_Conf._Ser._3143_012010.pdf
accesso aperto
Licenza:
Creative commons
Dimensione
1.56 MB
Formato
Adobe PDF
|
1.56 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.


