Deep learning‐based tomographic imaging of ECT for characterizing particle distribution in circulating fluidized bed

Abstract

The gas and solids in a circulating fluidized bed (CFB) are heterogeneously dispersed and a multiscale flow regime may form both in time and space. Accurate measurement of the fluidizing process is significant for investigating the multiscale gas–solid flow characteristics and the design, optimization, and control of CFBs in various applications. This article develops a deep learning‐based tomographic imaging of electrical capacitance tomography (ECT) to characterize the particle concentration distribution in a CFB. The deep tomographic imaging approach is realized through training a well‐designed convolutional neural network (CNN) with the numerically built dataset. Simulation results demonstrate that the average values of the relative image errors reconstructed by CNN in the test set are 0.1110 and 0.1114 for the 60 and 100 mm pipes, respectively, which are better than the average values of 0.1819 and 0.2519 by the Landweber algorithm. With the verification of the trained model based on the prepared data can image the unseen typical flow patterns better than Landweber, it is further used to investigate the particle flow characteristics of a lab‐scale CFB. Experimental results reveal that the developed deep tomographic imaging of ECT can successfully measure the fluidized particle distribution in both the 60 and 100 mm pipes, showing good prediction and generality of the designed CNN model. A flow regime transformation from “annular” flow to “core‐annular” flow and pneumatic conveying is observed under the tested conditions. Besides, the flow regime would be highly affected by the fluidized gas flow rate and the initial bed height.

Influence on Sample Determination for Deep Learning Electromagnetic Tomography

“…This is the change in sample size used in 21 randomly selected papers on DL-ET in recent years [ 16 , 22 , 23 , 24 , 25 , 26 , 27 , 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 , 38 , 39 , 40 , 41 ]. In Figure 1 , the size of each point represents the size of the sample set used in an article, while different colors are assigned to them for easy distinction.…”

“…Li et al generated a dataset, a relatively large sample set containing 100,000 data samples through numerical simulation. Each sample includes a normalized capacitance vector and a particle concentration distribution image [ 28 ].…”

A general neural network model co-driven by mechanism and data for the reliable design of gas–liquid T-junction microdevices

“…23,24 This description does not allow a good understanding of the internal principles and mechanisms of action, but due to its powerful fitting capability, the prediction of complex systems can be achieved. In chemical engineering, data modeling through machine learning is becoming increasingly popular, including research into reaction mechanisms, 25 hydrodynamic behavior, 26 and material design. 27 In the field of multiphase flow, machine learning is equally compelling.…”