LAUSR

Abstract As single component chemiluminescence cannot accurately measure the distribution of heat release rate, this paper put forward the idea of employing OH* and CH* chemiluminescence for the measurement. However, it is difficult to characterize the relationship between chemiluminescence and heat release rate, so the method of deep learning was applied to process numerical simulation results of methane-air steady premixed flames and a deep neural network model was developed to determine the distribution of heat release rate with OH* and CH* chemiluminescence. The evaluation indexes of the model performed satisfactorily: root mean square error of the normalized heat release rate is less than 0.13, mean relative error of peak and opening distance is below 3%, and mean error of peak position less than 0.01 mm. The validation results showed that OH* and CH* chemiluminescence could properly measure the distribution of heat release rate: relative error of peak ranges from −6% to 6%; on test dataset, correlation coefficient between predictive results and simulation results is above 0.99 in terms of heat release rate peak, peak positions and opening distance.