Data-driven optimal design of premixed hydrogen-air combustion in a microcombustor

Understanding micro-scale combustion mechanisms is vital for designing efficient micro-power generation devices, as these offer higher energy density than conventional batteries. The present study explores the microcombustion of premixed hydrogen-air by focusing on three critical parameters: inlet diameter ( D ), aspect ratio ( A R ), and Reynolds number ( R e ). We perform numerical simulations to investigate how these parameters influence maximum gas temperature ( T m a x ) and water production rate (MWPR), and thereby the combustion efficiency. Premixed hydrogen-air combustion in a chamber with adiabatic walls is represented by a reaction mechanism with 19 reversible elementary reactions and nine species. Comparing various reaction models reveals that species transport model for laminar flows provides best validation with previous simulations. Larger diameters (and lower R e ) generally lead to higher T m a x , while higher AR showed varied effects depending on the specific parameter range. The MWPR remains relatively constant with D , however, smaller chambers limit combustion efficiency, as reduced residence times do not allow gas mixtures sufficient time to complete the reaction. To model the nonlinear relationships between microcombustion outputs and the input design parameters, high-degree polynomials are required. The predictions from polynomial regression reveal a maximum relative error of 3% for MWPR and 8% for T m a x . The artificial neural network (ANN) provides a more reliable and generalized model with a maximum accuracy of 97.2% obtained using data from 92 simulations. For randomly chosen training data sets having 66%, 70%, 76%, and 80% training data, the accuracy from ANN is between 90%–98%. This demonstrates that the developed model generalizes well for different subsets of validation data. The polynomial regression and ANN serve as suitable alternatives for determining behavior of different input design parameters on hydrogen-air microcombustion, without performing expensive computations. • Species transport reaction model is best approach for premixed microcombustion of hydrogen-air. • Gas temperature and efficiency decreases due to incomplete combustion in smaller chambers. • Predictions from polynomial regression show a maximum relative error ≈ 3–8%. • Predictions from ANN are within ± 2 % of actual data for randomly chosen training data. • Polynomial regression and ANN provide optimal microcombustor design without expensive computations.