An Artificial Intelligence Regression Model for Prediction of NOx Emission from Flame Image

In this study, NOx emission has been estimated by processing the flame image of visible wavelength and its experimental verification has been presented. The experimental study has been performed by using a domestic coal boiler with a capacity of 85000 Kcal / h. The real NOx value has been measured from a flue gas analyzer device. The flame image has been taken by CCD camera from the observation hole on the side of the burner. The data set which is related to instantaneous combustion performance and flame images was recorded simultaneously on the same computer with time stamps once a second. The color flame image has been transformed into a gray scale. Features have been extracted from the gray image of flame. The features are extracted by using the cumulative projection vectors of row and column matrices. ANN regression model has been used as the learning model. The relationship between flame image and NOx emission has been obtained with the accuracy of R = 0.9522. Highly accurate measurement results show that the proposed NOx prediction model can be used in combustion monitor and control systems.