Skip to Main Content
Accurate and robust recognition of burning state for sintering process of rotary kiln plays an important role in the design of image-based intelligent control systems. Existing approaches such as consensus-based methods, temperature-based methods and image segmentation-based methods could not achieve satisfactory performance. This paper presents a flame image-based burning state recognition system using a set of heterogeneous features and fusion techniques. These features, i.e., the color feature, the global and local configuration features, are able to characterize different aspects of flame images, and they can be extracted from pixel values directly without segmentation efforts. In this study, ensemble learner models with four types of base classifiers and five fusion operators are examined with comprehensive comparisons. A total of 482 typical flame images, including 86 over-burning state images, 193 under-burning state images, and 203 normal-burning state images, were used in our experiments. These images were collected from the No. 3 rotary kiln at the Shanxi Aluminum Corporation in China, and labeled by the rotary kiln operational experts. Results demonstrate that our proposed image-based burning state recognition systems outperform other methods in terms of both recognition accuracy and robustness against the disturbance from smoke and dust inside the kiln.