Analysis of flame images in gas-fired furnaces.

摘要

Combustion is a key issue in gas-fired furnaces in various industries such as glass manufacturing. Its chemical reaction is based on two substances, oxidizer and fuel. Its quality depends on their composition, which are measured in terms of rate of flow and oxidizer to fuel (O/F) ratio by the furnace control system. Monitoring is crucial since improper composition produces hazardous byproducts and may waste expensive fuel.;This research proposes a promising system architecture that provides a method for assessing combustion quality by analyzing two-dimensional furnace flame image and correlates it with its fuel and oxidizer composition as reflected by the furnace control reading. The approach utilizes both image processing and machine learning techniques integrated with artificial intelligence techniques to identify correlations between flame characteristics and fuel flow rate and O/F ratio. Its conceptual design, implementation and evaluation are executed based on a set of experimental runs sampled at nine different composition of fuel and oxidizer flow rates taken from a pilot-scaled glass furnace.;A color CCD camera is used for capturing the furnace flame images. The images are processed using image processing techniques, from de-interlacing, cropping, image segmentation using Otsu's thresholding and image enhancement using proposed intensity suppression. Nine features are used to quantify the flame condition of which four are uniquely introduced in this study. Feature selection process is utilized to identify key features for the classification using wrapper method and decision tree classifiers. Fuzzy logic is then introduced to provide capability in classifying fuel level and O/F ratio beyond the known test data. Membership functions are designed and modeled based on key features output distribution, using generalized bell curve shape with parameters obtained by curve fitting and cubic interpolation technique.;The final architecture is implemented, tested and proven capable to provide insight into the combustion quality in term of its fuel and O/F ratio class within seconds.