Energies, Vol. 18, Pages 5730: Machine Learning Improves Performance Prediction and Interpretation of Efficiency Influencing Factors of a Novel Basalt-Fiber-Bundle Thermal Flow-Reversal Reactor for Methane Recovery

Energies doi: 10.3390/en18215730

Authors:
Rao Kuang
Bin Du
Peter D. Lund
Jun Wang
Yanying Liu

Low-concentration methane emissions from mines can be recovered using different reactor designs. Here, different artificial intelligence network techniques were employed to predict thermal performance of a basalt-fiber-bundle thermal flow-reversal reactor and investigate the influence of input parameters. The Back Propagation (BP) model gave the best accuracy (R2 = 0.974 for outlet temperature, 0.967 for thermal efficiency), exceeding that of traditional Computational Fluid Dynamics (CFD) simulations. For the present design, when flow velocity exceeded 1.5 m/s, the outlet gas temperature shifted from rising to falling, explained by the heat transfer between the gas and the solid inside the flow channel. Increasing the length of the flow-reversal period in the high-temperature phase reduced the outlet temperature, e.g., an increase from 60 s to 200 s decreased the outlet temperature by 34.1 K. Increasing inlet methane concentration (e.g., from 0.3% to 0.8%) first showed a slight improvement in thermal efficiency but further increase accelerated the oxidation reaction rate inside the reactor, reducing the temperature difference between the solid and gas in the channel, which slowed the heat exchange process and resulted in a downward trend in efficiency. The results indicate that the reactor can handle a wide range of exhaust gas concentrations, being suitable to treat low-methane-concentration exhaust gas. The BP model helped to establish the theoretical basis for setting optimal parameters values for the operation of the proposed reactor.

Source link

Rao Kuang www.mdpi.com