Monitoring Flames in an Industrial Boiler Using Multivariate Image Analysis Honglu Yu and John F. MacGregor McMaster Advanced Control Consortium, Dept. of Chemical Engineering, McMaster University, Hamilton, Ontario L8S 4L7, Canada DOI 10.1002/aic.10164 Published online in Wiley InterScience (www.interscience.wiley.com). An on-line digital imaging system is developed for monitoring flames in an industrial boiler system. The information extracted from the RGB flame images is used to predict the performance of the boiler system. A practical method based on multivariate image analysis techniques and partial least squares is developed to efficiently extract informa- tion from the rapidly time varying flame images, and to predict boiler performance, NO x and SO 2 concentration in the off-gas, and the energy content of an incoming waste fuel stream. The approach is very general and can be applied to a wide range of combustion processes. © 2004 American Institute of Chemical Engineers AIChE J, 50: 1474 –1483, 2004 Keywords: flames, boilers, image analysis, performance monitoring, principa-component analysis, partial least squares Introduction Combustion plays an important role in many industrial pro- cesses. The efficiency of the combustion usually has a great influence on the economics of the process and on its environ- mental impact. In combustion processes fuel and oxidizer (typically air) are mixed and burned. Generally, two categories can be identified based on whether the fuel and oxidizer are mixed first and then burned (premixed) or whether combustion and mixing occurs simultaneously (nonpremixed). Each of these categories is fur- ther subdivided based on whether the fluid flow is laminar or turbulent. Among the four classes, turbulent nonpremixed com- bustion processes are of interest in many industrial applica- tions. They appear in jet engines, diesel engines, steam boilers, and furnaces, for example. The use of the nonpremixed com- bustion is widespread because it is safer to handle than pre- mixed combustions. However, nonpremixed combustions in- clude more complex chemistry and are harder to control. Unless very sophisticated mixing techniques are used, nonpre- mixed flames show a yellow luminescence, caused by glowing soot particles. The colors of the flames indicate the combustion region and the temperature of the field. This latter feature allows for the possibility of using color flame images to mon- itor the combustion process. Visualization methods have been used to study combustion flames in laboratories. Shimoda et al. (1990) reported a com- bustion diagnostic system where the radiation energy and tem- perature profiles of flames were quantitatively identified from the flame images and the concentrations of unburned carbon and NO x in the exhaust gas were estimated in a coal-fired boiler. Yamaguchi et al. (1997) developed a fiber-optical im- aging sensor to detect the air/fuel ratio in a premixed-type gas-fired model combustor by monitoring the radiant intensities of flames over three spectral bands. Huang et al. (1999) set up a flame-flicker monitoring system, where the flicker of a gas- eous flame was quantified by computing the oscillation of the radiant intensity of individual pixels within the luminous re- gion of flame images. Lu et al. (2000) designed and evaluated an instrument system for monitoring, characterization, and evaluation of fossil-fuel–fired flames in a utility boiler. Geo- metrical and luminous parameters of the flame were deter- mined from the images. Wang et al. (2002) reported a method to predict NO x emissive concentration for a coal boiler by using color flame images and neural networks. However, few of the above studies used turbulent nonpremixed flames as objects. In many industrial furnace and boiler systems, television systems have been installed to monitor the flame. However, most of the time the only information the flame images are providing is whether the flame is burning. Because of the Correspondence concerning this article should be addressed to J. F. MacGregor at macgreg@mcmaster.ca. © 2004 American Institute of Chemical Engineers 1474 AIChE Journal July 2004 Vol. 50, No. 7