Data-Driven Fluidized Bed Flow Field Reconstruction Using Limited Measurements

Abstract

The rapid acquisition of internal flow field data in fluidized beds is essential for applications in monitoring, prediction, risk warning, prevention, and diagnostics. This study proposes an effective approach for reconstructing the solid volume fraction field in fluidized beds based on sensor placement optimization. First, proper orthogonal decomposition (POD) is applied to the training set to reduce the dimensionality. Next, different sensor schemes are discussed, with Gappy POD utilizing sensor selection based on QR decomposition with column pivoting, reducing reconstruction errors from 671.3 and 541.0 to 58.1%, compared to random and regular sensor selection schemes. Furthermore, when the number and spatial distribution of sensors are fixed, multilayer perceptron (MLP) models deliver the best reconstruction performance, reducing errors by approximately 9%. These findings suggest that the QR sensor scheme can effectively guide sensor placement while MLP models can be employed to further optimize reconstruction accuracy.