The combustion quality determination in power station boilers is of great importance so as to avoid air pollution. The emission of harmful gases as a result of incomplete or partial combustion can be reduced by monitoring the flame images at the furnace of the boiler. Complete combustion minimises the exit of NOx, SOx, CO and CO2 emissions also ensuring the consistency in load generation. If the combustion was complete then the flue gas emissions are maintained within minimum limits. Flame image analysis was done using Support Vector Machine (SVM) and classification based on the combustion quality was done with Meta Classification via Clustering, Cross Validation Parameter Selection (CVP), Radial Basis Function Network (RBFN) and MultiLayer Perceptron (MLP). Various performance measures are used for cross validation to estimate the combustion conditions. This research work is a combination of Fisher’s linear discriminant analysis and radial basis function method for identifying the combustion conditions from the flame images of a boiler. The images in the control room are acquired using an infrared camera fixed to the inner portion of the boiler. The features of the image are further extracted using correlation. The dimensions of the input for the training patterns are reduced from 30 to 2 using optimal discriminant plane technique. Two projection vectors 1 and 2 is calculated for reducing the dimension of the input pattern. During training and testing of Radial Basis function Network (RBF), the number of input nodes is 2. Nineteen patterns have been used for training RBF and another 19 patterns for testing the RBF. Results obtained are promising and positive to implement for closed loop monitoring of the boiler.

Combustion quality, Support vector machine, Radial basis function network, Multilayer perceptron, Meta Classification via Clustering, CV Parameter Selection, intelligent technique, Automation