讲座时间：2023年12月13日（星期三）14:00-16:00

讲座地点：河北工业大学会议服务中心第二会议室

讲座题目：State Estimation and Neural Network Approaches for Two-Phase Flow Imaging using Dual-Modal Electrical Tomography

讲座嘉宾：Prof. Marko Vauhkonen

Marko Vauhkonen received his PhD in physics in 1997 at the University of Kuopio, Finland. After graduation, he worked as a researcher and research director at the same university, until moving to Germany in 2006. There he worked as a Marie-Curie Research Fellow for two years at the Philips Research GmbH, Aachen. During 2008–2009 Vauhkonen worked as a CTO in a spin-off company Numcore Ltd. until starting as a professor at the University of Kuopio (currently University of Eastern Finland), Department of Applied Physics (currently Department of Technical Physics), in 2009. His research interests include inverse problems, time-varying reconstruction, process tomography and medical imaging such as PET, SPECT and MRI. He has published more than 120 scientific journal articles.

This talk shows results of a study, in which we explore the potential of state estimation and neural networks for image reconstruction in dual-modal tomography of two-phase oil-water flow. The accurate measurement of two-phase flow quantities is crucial for effective production management across various industries. However, the complexity inherent in two-phase flow poses challenges in accurately estimating the quantities, necessitating the development of reliable reconstruction techniques. Our approach involves utilizing electromagnetic flow tomography (EMFT) for velocity field estimation and electrical tomography (ET) for determining phase fraction distributions. To account for the contribution of the velocity field to the temporal evolution of the phase fraction distribution, we employ a convection-diffusion model in the state estimation process. Furthermore, the extended Kalman filter (EKF) and fixed-interval Kalman smoother (FIKS) are used for reconstructing the spatio-temporal velocity and phase fraction distributions. In addition, we study the possibility of utilizing deep neural networks (DNNs) for directly estimating the oil flow rate in two-phase oil-water flows. The effectiveness of our approaches is demonstrated through simulations and experimental investigations on a laboratory setup, considering different cases with stationary and non-stationary average flow speeds. The simulations and experimental results reveal that the proposed time-varying approach outperforms conventional stationary reconstructions, highlighting the potential of state estimation for achieving more accurate dynamic image reconstruction in dual-modal tomography of two-phase flow. Furthermore, the simulation results demonstrate promising potential of the proposed DNN approach in accurately estimating the oil flow rate in complex two-phase flow systems.