Prediction of Oxygen Content using Deep Learning CNN Architecture for the Classification of Blast Furnace Gas Fired Boiler Images

Abstract

Prediction of oxygen content in a combustion process is one of the prime and arduous task. As high temperatures are involved, there are seldom any equipment available, that can be placed inside the furnace to make measurements. to obliviate this problem, in this work, a convolution neural network (CNN) is applied to an industrial process of furnace combustion. Flame images of a working gas fired boiler are obtained by a high definition camera and an artificial neural network, CNN is applied to anticipate the oxygen content present in flue gas of a BFG gas fired boiler. A multilayer CNN model is used to describe better, the key patterns in a combustion process by extracting the nonlinear aspects. Using a CNN model and a multilayer representation of the CCD flame pictures, more in-sightful data regarding the physical characteristics of flames can be defined. This concept is applied to flame images obtained on-site from a real combustion system. The loss obtained is 0.04, quite a low value, after the model training, and acquired 97% accuracy, which is very good for classification tasks during testing.