Application if machine learning methods for diagnostics and control of coal dust combustion modes

The study examined the possibility of monitoring and identifying anomalies in the combustion process based on the patterns of flame visualization in the combustion space. Experimental studies were carried out on a semi-industrial stand to study the combustion of pulverized coal. Combustion images were classified using a deep learning algorithm for a finite number of corresponding excess air coefficients (α). Based on the excess air coefficient, we divided combustion modes into three main classes: efficient, ineffective and anomalous. This type of classification provides a simple approach to optimize the combustion process and alert the system in the event of critical events. The resulting model classified combustion modes with an accuracy in the range of 87–91 %.