Multi-source data driven-based rotary kiln burning state recognition using heterogeneous features and fuzzy integral

Accurate and robust recognition of burning state for alumina rotary kiln sintering process plays an important role in the design of intelligent control systems. Existing approaches such as image segmentation-based methods could not achieve satisfactory performance. This paper presents a novel multisource data driven-based burning state recognition model to further improve our existing flame image feature-based recognition result. Four heterogeneous features, i.e. flame image ROIs color, global, and local configuration features, and process variable feature, are able to comprehensively characterize different aspects of burning state, and the flame image-based features can be directly extracted without segmentation efforts. In this study, pattern classifier and fuzzy integral operator are examined with comprehensive comparisons. A total of 482 typical flame images labelled by the rotary kiln operational experts, including 86 over-burning images, 193 under-burning images, and 203 normal-burning images, and associated process variable at the same moment from No. 3 rotary kiln at Shanxi Aluminum Corp were used in our experiments. Results demonstrate that our proposed multi-source data driven-based burning state recognition model outperforms individual feature-based methods and other recognition methods in terms of both recognition accuracy and robustness.