Flame imaging as a diagnostic tool for industrial combustion

Abstract

Different approaches for condition monitoring of combustion systems based on the capture and processing of flame images are presented and discussed. The objective is to devise methods capable of converting geometrical and luminous data into reliable information on the state of practical combustion systems. Suitable processing methods are needed to extract representative information from flame images, as they contain large amounts of data whose physical interpretation is not, in general, straightforward. A thorough experimental programme has been conducted in a model industrial burner, in order to create a database of flame images for subsequent analysis. One of the options is the extraction and analysis of representative image features; even though the parametric studies revealed good sensitivity to changes in combustion regimes, this approach may entail some information loss and requires adaptation to the specific system. The other two methods explored are designed as flame identification techniques based on the whole image. One of them use self-organising feature maps and yields as output the most probable combustion regime, among those previously characterised. The other one is an adaptation of a speech recognition method and informs about the probability of an unknown state to correspond with the different combustion regimes. Their performance was tested in terms of their success rate for identification as well as on their capabilities to estimate NO x emission, as a representative outcome of the combustion process. Results in other combustion situations (oil burner and premixed combustor) are also reported. The good results obtained in all cases are thought to support the potential of the methods described for flame monitoring.