基于 k-means 机器学习方法的气固循环流化床颗粒聚团特性

Circulating fluidized bed is widely used in industrial production due to its good gas-solids mixing. Clustering characteristics affect the gas-solids contacting and therefore the heat/mass transfer and the yield of the products as well as the selectivity. To get more detailed information effectively, a high-speed camera was used to visualize the flow field in a two-dimensional circulating fluidized bed at the superficial gas velocity, U_g, equal to 5—9m/s, and the solids circulating rate, G_s, of 50—300kg/(m²·s). The k-means machine learning algorithm was then utilized to assist the image processing to identify the cluster effectively and quantificationally. Results showed that as G_s increased from 50kg/(m²·s) to 300kg/(m²·s) at U_g=9m/s, the cluster frequency almost doubled from 116Hz to 327Hz. The average cluster concentration was correlated with the local lateral position. The average cluster concentration was more uniformly distributed in the central region (y/Y from 0 to 0.7). Toward the near wall region (y/Y from 0.7 to 0.9) it increases rapidly. The change of the average cluster concentration near the wall was nearly three times higher than in the central region. Both the average cluster velocity and the average cluster equivalent diameter displayed a similar trend, decreasing from the center toward the wall in the lateral direction. Quantitative prediction equations for cluster parameters were obtained based on the experimental data. The relative errors were all within 30%. Cluster characteristics were studied quantitatively and systematically in this study. These results can support the development of a gas-solids flow model and process intensification for circulating fluidized beds.