Optimization of sustainable biogas valorization to hydrogen via tri-reforming process in packed bed membrane reactor: An integrated CFD-ML digital twin approach

The integration of H2 production with greenhouse gas mitigation offers a promising pathway for sustainable energy generation. This study focuses on optimizing a Pd-Ag packed bed membrane reactor (PBMR) for biogas tri-reforming to produce H2, employing a digital twin framework that combines computational fluid dynamics (CFD) and machine learning (ML). Indeed, industrially, H2 generation is performed by steam reforming of natural gas, while the adoption of biogas as a renewable source may contribute to mitigate the greenhouse gases emission in the atmosphere. Furthermore, rather than the unique steam reforming reaction commonly adopted to produce H2, biogas tri-reforming integrates dry reforming, steam reforming, and partial oxidation reforming, with the scope of transforming CH4 and CO2 contained in a biogas stream into H2. Therefore, to assess the efficiency of the tri-reforming process approach, it was theoretically compared to biogas steam reforming. CFD simulations were performed to evaluate the performance of PBMRs in comparison to packed bed reactors (PBRs), with experimental model validation conducted to ensure accuracy. Four ML models Gaussian Process Regression (GPR), Artificial Neural Networks (ANN), Random Forest Regression (RFR), and Support Vector Regression (SVR) were developed to create predictive models and optimize performance parameters. Key findings revealed that PBMRs consistently outperformed PBRs due to process intensification enabled by the Pd-Ag membrane. Among the ML models, the GPR model demonstrated near-perfect accuracy, while ANN and SVR also showed high predictive capabilities, with R2 values of 0.987 and 0.98, respectively. Bayesian optimization identified the optimal operating conditions for tri-reforming in PBMRs as 773 K, an H2O/CH4 ratio of 2.84, and a WHSV of 0.1 h−1, where the simulations revealed the best performance such as CH4 and CO2 conversion equal to around 97 % in both cases, H2 yield to 81 %, H2 recovery to 58 %, and H2 selectivity equal to 87 %, aligning with similar results obtained by CFD analysis. Moreover, the predictive equations underscored that while temperature was critical for enhancing reaction conversions, H2O/CH4 ratio and WHSV showed a more substantial and nuanced influence on both feed conversion and H2 selectivity. The digital twin approach significantly reduced computational costs, enhanced reactor dynamic insights, and provided a robust tool for optimizing sustainable processes, highlighting how the PBMR may improve by 20–25 % H2 selectivity and achieve a marked reduction in CO2 emissions, aligning with sustainable energy goals.