Register or Log In

International Society for Industrial Process Tomography

12th World Congress on Industrial Process Tomography

An ECT image reconstruction method of the Tiny tube based on Deeplab-v3 Network

Y.X. Han1, Z.Y. Liu1, Y.Z. Zhang1, J.T. Sun1*, L.J. Xu1

1 School of Instrumentation and Optoelectronic Engineering, Beihang University, Beijing, China

*Email: jiangtao_sun@buaa.edu.cn


ABSTRACT

Gas-liquid two-phase flow detection in tiny tubes holds significant importance for chemical engineering and aerospace applications. Electrical Capacitance Tomography (ECT) has emerged as a prominent technique for multiphase flow measurement due to its non-invasive nature, cost-effectiveness, and structural simplicity. This study establishes a 2D simulation model for ECT data acquisition in tiny tubes (6mm inner diameter, 9mm outer diameter, 1.5mm glass walls), with nitrogen gas (ε=1) and deionized water (ε=81) as the working media. To address the limitations of conventional ECT algorithms in accuracy and computational efficiency, we propose a 6-electrode ECT system coupled with a novel DeepLab-v3-based reconstruction network featuring an encoder-decoder architecture. The encoder employs a structurally optimized ResNet101 backbone, where the conv4_x blocks are strategically compressed to reduce computational load while preserving feature extraction capability. Multiscale contextual information is captured through an Atrous Spatial Pyramid Pooling (ASPP) module, which integrates parallel convolutional branches with varying dilation rates (r=2,3,6). The decoder synthesizes high-level semantic features (5×5 resolution) with low-level spatial details (17×17 resolution) through channel-wise concatenation and progressive upsampling, ultimately reconstructing 67×67 permittivity distribution maps. This architecture achieves real-time performance at 20 frames per second with a 13.84% relative error on test data.


Keywords: Comsol Simulation, DeepLab-v3 Network, Electrical Capacitance Tomography, tiny-tube

Industrial Application: Gas-liquid two-phase flow detection

Sign-in to access the full text