Abstract
Electrical impedance tomography (EIT) is a nondestructive testing technique, which has great potential to be used for the detection of the cross-sectional gas-holdup ratio (CGR) of gas–liquid two-phase flow. Due to the nonlinear and ill-posed characteristics of the EIT image reconstruction, the accuracy of image-based CGR estimation methods is poor. To improve the detection accuracy, a learning-based direct CGR estimation method is proposed in this study, which applies a novel multiscale attention network (MSA-Net) to directly estimate the CGR from the voltage measurements. Multiscale feature extraction and residual structure are introduced into MSA-Net to fully extract the features, and attention unit (AU) is also used to capture high-frequency features. As a result, accurate and robust CGR estimation can be realized by MSA-Net. ResNet18 and single-scale attention network (SSA-Net) are selected for comparison. The simulation results indicate that the relative error (RE <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$_{\text {CG}}{)}$ </tex-math></inline-formula> of CGR estimated by the learning-based direct CGR estimation method is far lower than that of the image-based CGR estimation methods. Compared with ResNet18 and SSA-Net, the RECG of the CGR estimated by MSA-Net is lower, and the range is 0.07%–0.36%. Moreover, MSA-Net shows good noise robustness simultaneously. The experiment is also set to test MSA-Net. The range of RECG is 0.1%–1.12%, which further verifies the practicability of the proposed learning-based direct CGR estimation method.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call