Carl Duchesne, a professor of chemical engineering at Université Laval in Quebec City, and graduate student Gerard-Silard Szatvanyi have developed a method that promises to get around measurement problems associated with taking the temperature of materials in a rotary kiln. They take a color digital picture of the flame and subject it to multivariate image analysis.Multivariate regression analysis of images acquired of the flames within a rotary-style kiln helps predict discharge temperatures without having to access the oven’s interior or the material being processed. Within a flame image (A), flames with similar coloration can be analyzed even when flickering causes the points of the flame to appear at different positions within the image. This enables the creation of score-density histograms (B). Courtesy of the American Chemical Society.Rotary kilns have a number of advantages for industrial processes: They are versatile, enabling the drying of products as varied as fish meal, minerals and sawdust; they can calcinate lime and coke, heating the substances to a high temperature below the melting point so that water and CO2 are driven off; and they can roast and sinter ore.Kilns suffer from some problems, however, including inefficiency. One reason is that kiln operators typically overheat them. The operators have a good motivation for this, Duchesne noted, in that they ensure that the discharge temperature exceeds a certain lower limit to meet a particular quality specification.Inside the kiln is a complex, dynamic system of chemical reactions and of materials in various phases. It is difficult to model, and there is variability in the solid discharge temperature. Existing measurement techniques have a hard time determining the temperature of solids within a rotating kiln. To make sure that they hit the requirements, operators thus overshoot by applying more heat than is necessary. The cost of doing so may be several hundred thousand dollars per year in added fuel used, and burning that extra fuel also adds to the pollution produced.Flame images, such as those that Duchesne and Szatvanyi are employing, have been used by others for similar measurements, although not on kilns. Those efforts have met with limited success. However, they were performed in gray scale and so suffered from not capturing a complete picture.“We have shown that using a very efficient method to extract the spectral — that is, the RGB or multivariate image — information, one can get more process-relevant information,” Duchesne said.In a demonstration that they reported in the June 21 issue of Industrial & Engineering Chemistry Research, the researchers mounted a JVC 640 × 480-pixel CCD camera behind the burner of a 220-ft-long, 12-ft-diameter kiln operated by QIT-Fer et Titane Inc. of Sorel-Tracy, Quebec, Canada. For each pixel, they captured red, green and blue intensity, for a total of 24 bits per pixel. While the kiln was operating in production, they imaged a frame every 10 seconds for months at a time over nearly a year, collecting more than 80,000 images.They analyzed the images, extracting flame features using multivariate image analysis. The benefit of this approach was that it classified pixels according to spectral characteristics and not to location. It was not necessary to first locate the flame, which bounced around from frame to frame as a result of turbulence, and that saved computing time. They then correlated the multivariate analysis to the solids discharge temperature, using a small part of the collected images to build a model and the rest to validate it. They found that 80 percent of the discharge temperature variations could be explained by the color.“The flame itself is not important for predicting discharge temperature,” Duchesne said, “but the color of the kiln’s walls and the solids were the most important features used by the model. This makes a lot of sense from a heat-transfer perspective.”Applications include quality control schemes for rotary kilns and other combustion systems. According to Duchesne, a few companies already have expressed interest in the technique.Contact: Carl Duchesne, Université Laval, Quebec City; +1 (418) 656-5184; e-mail: carl.duchesne@gch.ulaval.ca.