Investigation of ML algorithms for prediction of CFD data of fluid flow inside a packed-bed reactor
This study considered integrated simulation of a chemical reactor for the production of hydrogen from steam reforming of methanol via a hybrid approach. Computational simulation was carried out via CFD (computational fluid dynamics) to determine H2 concentration in the reactor, and the concentration...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-06-01
|
| Series: | Case Studies in Thermal Engineering |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2214157X25003533 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | This study considered integrated simulation of a chemical reactor for the production of hydrogen from steam reforming of methanol via a hybrid approach. Computational simulation was carried out via CFD (computational fluid dynamics) to determine H2 concentration in the reactor, and the concentration data was then used for machine learning (ML) modeling. For CFD, two-dimensional mathematical modelling and simulation were carried out for the investigation of mass transfer and hydrodynamics of methanol steam reforming in a packed-bed reactor. For ML modeling, Decision Tree (DT), Multi-Layer Perceptron (MLP), and Polynomial Regression (PR) were used in estimating hydrogen concentration using r(m) and z(m) as predictors. Hyper-parameter optimization was conducted employing Sequential Model-Based Optimization (SMBO) considering K-fold as the objective metric to optimize models. The dataset was partitioned into training and testing sets in an 80-20 proportion. The MLP model demonstrated superior performance with a 5-Fold Mean (R2) of 0.997750 and the lowest error rates. The DT model also performed well, achieving a 5-Fold Mean (R2) of 0.991316 and slightly higher error rates. Conversely, the PR model exhibited notably inferior performance, displaying a 5-Fold Mean (R2) of 0.880576 and elevated error. These outcomes highlight the MLP model as the most proficient in forecasting hydrogen concentration within this scenario. It was observed that all methanol can convert to the products at 573 K and P = 1 atm. The hydrogen percentage was about 67 % at the outlet of the packed-bed reactor. Increasing the temperature from 473 to 573 K increased methanol conversion from 11.17 % to 100 % in the packed-bed reactor. These findings emphasize the significance of meticulous model selection and hyperparameter tuning to attain heightened predictive precision in machine learning endeavors. |
|---|---|
| ISSN: | 2214-157X |