Cluster Identification by a k-means Algorithm-Assisted Imaging Method in a Laboratory-Scale Circulating Fluidized Bed

Particle clusters for FCC particles in a gas-solid circulating fluidized bed with a 12.4 m high riser and a 5 m high downer were identified from the images of the gas-solid flow by a k-means machine learning algorithm-assisted processing method. An optimal k value of 3 was determined and justified by several evaluation criteria for the k-means algorithm. The solid holdup obtained from the processed images agrees well with that from the optical fiber method. The particle cluster characteristics between the riser and downer, such as the cluster solid holdup, equivalent diameter, velocity, and frequency, were extracted from the processed images and then compared in detail for the first time. The cluster solid holdup and the cluster velocity in the riser (epsilon(cl) = 0.05-0.20, V-cl = 4-10 m/s) are much higher than those in the downer (epsilon(cl) = 0.005 0.020, V-cl = 2-5 m/s). The cluster equivalent diameter and the cluster frequency in the riser and downer are similar (d(cl) = 2-10 mm, f(cl) = 100-400 Hz). Empirical correlations of the cluster characteristics with the local flow conditions and the operating parameters in both the riser and downer are further studied.

Automated summary

Particle clusters for FCC particles in a gas-solid circulating fluidized bed were identified and characterized using a machine learning algorithm-assisted processing method. The characteristics of the particle clusters were analyzed and compared between the riser and downer sections, revealing significant differences in their solid holdup and velocity. The study also investigated the correlations between the cluster characteristics and local flow conditions and operating parameters.