Recognition, Retrieval And Clustering Of The Flame Image In Alumina Sintering Rotary Kiln

In alumina sintering process, there exist some problems such as complicated procedure, high energy consumption and low information level. The research of novel sintering process and procedure informationization technology have been an important path to improve automation level, save energy consumption and prolong the lifetime of the sintering kiln.This dissertation, financial supported by the National Natural Science Foundation of China, selected the alumina rotary kiln and flame image as objects and studied the sintering process and informationization of the rotary kiln. Firstly, a novel technology for alumina rotary kiln sintering and their numerical simulation were investigated. Then the rotary kiln flame image were analyzed, and the flame image features calculation, flame image recognition, flame image retrieval and clustering mining were also discussed. The major conclusions and results were descriobed as follows:(1) A new technology for alumina rotary kiln sintering was suggested, namely the spray dryer sintering. Its equipment and process flow were disscussed. And a 3D numerical model of the key equipment drying tower was built with Fluent software. The flow field distributions inside the tower were analyzed. The results indicated that the distributions of the temperature and velocity were reasonable, and the drying speed was satisfactory.(2) A flame image segmentation method based on multi-region was designed. The segmentation experiments results indicated that the proposed method could achieve the same result of the FCM algorithm with the speed of single threshold segmentations. Moreover, several features of the flame images were calculated: 5 general features of flame and clinker, 1 flame shape descriptor based on the detailed analysis of flame shape, 4 textual features and 5 fractal dimensions features. At last, a flame image database was designed based on these images and the corresponding features.(3) Using textual, fractal and fire & clinker features and the artificial neural network, a rotary kiln flame image recognition method was investigated. The structure of the network was designed. And the training and testing samples were selected according to the expertise of the workers. The best network structure was determined through the recognition experiments. Finally, the results indicated that the optimal recognition rate of neural network was 92.3077%, which was satisfactory for industrial requirement.(4) Based on flame image recognition and pattern classification, a real-time simulation structure was proposed. Firstly, the Fluent simulation results database was constructed; secondly, the actual running state was identified based on flame image recognition and pattern classification; thirdly, the mathematical model between the simulation results and input parameters was built; and finally, the secondary simulation results were calculated. The secondary simulation experiment of the rotary kiln temperature distribution were performed, and the results showed that it could get satisfactory simulation result in very short time, which provided a new view for the temperature supervision of rotary kiln and could be easily expanded to other field.(5) The content-based flame image retrieval was studied. Through the similarity calculation of the low level features, it acquired a serial of similar retrieved images. Meanwhile, a rotary kiln flame image semantic extraction model and retrieval model were also proposed. The experiments results showed that it could not only effectively extract semantic features from rotary kiln image, but also returned good retrieval results.(6) A flame image clustering method based on image processing and gray association degree was designed, namely the Gry-k-Means clustering model. The rotary kiln flame image gray association degrees were calculated, and transformed into the gray weights which were applied in distance calculation; using this distance measurement, a new clustering algorithm was proposed based on the traditional k-Means algorithm. The new clustering algorithm was applied in the clustering analysis of rotary kiln flame image. The experiments with production data were carried out, and the results showed that the Gry-k-Means algorithm could provide improved results.(7) The rotary kiln flame image information system(RKFIIS) was finally designed based on the above research results. It integrated the flame image management, processing, recognition, retrieval and clustering. Furthermore, it had friendly GUI and could provide strong support for rotary kiln control and management.