Generative Adversarial Network Model for Two-Phase Flow Imaging by Electrical Impedance Tomography

Electrical impedance tomography (EIT) is a powerful tool for imaging two-phase flow systems in various applications; however, the reconstruction quality of EIT is often limited by the nonlinear and ill-posed nature of the inverse problem. In order to improve the quality of EIT reconstruction, a new method using a generative adversarial network (GAN) model is proposed for two-phase flow systems. The GAN model generator and discriminator are adversarially trained on boundary voltages and images of the target location. This training process allows the GAN model to learn the underlying patterns and features present in the EIT data, leading to improved reconstruction outcomes. Numerical and experimental studies are done for two-phase flow setup verification. Results show that the proposed GAN-based model manages to reconstruct target locations with a homogeneous background, indicating the potential for improving EIT-based two-phase flow imaging in industrial applications. Metrics are also compared with other deep learning methods, such as neural network (NN) and deep neural network (DNN), from which the proposed model overcomes image quality metrics such as structural similarity index (SSIM) and relative image error (RIE).