A novel method of efficiently using the experimental data for mechanism optimization: Theory and application to NH3/H2 combustion

In this work, a novel method of efficiently using the experimental data (EUED) is proposed to reduce the computational cost of evaluating the objective function. The EUED method involves splitting the full experimental dataset into several subsets that retain the essential features of the full dataset's effects on the influential reactions, with these subsets used in rotation during the iterations. The constraint probability density function (PDF) of the constraint frequency distribution spectrum of the influential reactions reflects the essential features of the experimental data's effects. Thus, the subsets should meet 2 criteria: first, the union of all subsets must equal the full dataset; second, the PDF of the frequency spectrum of the influential reactions in any subset should align with that of the full dataset. A strategy for allocating data into several subsets is proposed. An optimized NH₃ combustion model was developed using the EUED method. The prediction errors are 1.22 for species concentrations during pyrolysis, 1.67 for ST-IDT, 1.45 for species concentrations during oxidation, 1.84 for LBV, and 4.29 for RCM-IDT measurements, respectively. The 200 ST-IDT measurements, 911 LBV measurements and 172 RCM-IDT measurements are split into 4, 10 and 4 subsets, respectively. This approach reduces the computational costs of evaluating the objective function at each iteration by about 80 % during the NH₃ model optimization. The roles of the unimolecular decomposition reactions of NH₃, 2 H-abstraction reactions of NH_i by H and 4 reactions involving NH_i in NH₃ pyrolysis were discussed in detail. The optimization automatically weighs the rate constants of the 7 important reactions in an extraordinarily tangled and complicated reaction network, leading to satisfactory predictions of the NH₃, NH₂ and NH profiles.

Novelty and Significance Statement

In this work, a novel method of efficiently using the experimental data (EUED) is proposed to reduce the computational cost of evaluating the objective function. The idea of the EUED method is that the full experimental dataset is split into several subsets which remain the essential feature of the effects of full experimental data on the influential reactions, and the several subsets are used in rotation in the iteration. An optimized NH₃ combustion model was obtained using the EUED method with reducing 80 % computational costs of evaluating the objective function. The optimized NH₃ model outperforms the initial one and the models considered in this work.