Measurement of heat release rate using regression of chemiluminescence signals in premixed ammonia-methane/hydrogen-air flames

• Non-intrusive HRR measurement via chemiluminescence regression. • Gaussian Process Regression applied to NH 3 -CH 4 /H 2 -air laminar and turbulent flames. • HRR predicted accurately without known operating conditions. • Two excited species suffice for high-accuracy HRR prediction in ammonia flames. • Method enables future real-time monitoring of practical ammonia combustion systems. This paper proposes a non-intrusive method for quantitatively measuring heat release rate (HRR) through regression of chemiluminescence signals. The method is demonstrated using premixed ammonia-methane-air and ammonia-hydrogen-air counterflow twin laminar flames, with chemiluminescence data previously collected across a wide range of ammonia fractions, equivalence ratios, and strain rates. Opposed-flow 1-D simulations are employed to determine the true HRR of these twin flames, providing the complete dataset necessary to train an algorithm that correlates HRR with chemiluminescence intensities via Gaussian Process Regression (GPR). The accuracy of HRR predictions from the trained algorithms is evaluated using various input combinations, from including all recorded excited species (such as NO*, OH*, NH*, CN*, CO 2 *, and CH* for ammonia-methane-air flames, and NO*, OH*, NH* along with violet, blue, green, yellow, orange , and red colors for ammonia-hydrogen-air flames) to only one pair of species. Results indicate that UV-visible chemiluminescence alone can reliably infer HRR with high accuracy, better than 6.7% for ammonia-methane and 7.5% for ammonia-hydrogen flames, when all species are considered in the GPR, even if ammonia fraction, equivalence ratio, and strain rate are unknown. Further data show that chemiluminescence intensities from just two excited species, specifically OH*-CN* for ammonia-methane-air and OH*-NH* for ammonia-hydrogen-air, can provide average accuracies of 2.8% and 4.7%, respectively. The method is also applied to ammonia-hydrogen-nitrogen-air swirling turbulent flames, which better represent practical scenarios, using a new dataset of NH*, OH*, and NO* chemiluminescence intensities. It is demonstrated that, with proper data processing, this approach can accurately predict HRR for these more complex swirling turbulent flames.