Adaptive Edge-Weighted Graph Learning Network Architecture for Accurate ERT Image Reconstruction in Multiphase Flow

Sathesh Ammaiappan¹, Awais Ahmed¹, Guanghui Liang^1*, Feng Dong^1,2

¹Tianjin Key Laboratory of Process Measurement and Control, School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China
²School of Information and Intelligent Engineering, Tianjin Renai College, Tianjin 301636, China

^*Email: ghliang@tju.edu.cn

ABSTRACT

Electrical resistance tomography (ERT) uses a network of boundary electrodes to visualize and analyse processes involving multiphase flow effectively. The nonlinearity and ill-posed problems are challenging for ERT reconstruction. However, reconstruction image accuracy is in demand. Due to a lack of suitable training, the existing deep learning network for ERT image reconstruction depends on sparse information flow and gradient flow. The traditional reconstruction algorithm challenges the ability to capture the spatial and structural topological relationship among internal conductivity features. The challenges can be encoded through the non-Euclidean structure of the ERT measurement, propagating feature representation between topological relationships. Capturing spatial and topological dependencies requires adaptive learning to handle irregular prominent feature positions and conductivity domains. Therefore, a novel adaptive graph-based convolutional neural network is proposed to achieve higher imaging accuracy. The dynamic graph learning through edge weights jointly with node features using the RMSprop optimization method. The adaptive edge learning can identify object geometries and real-time flow patterns. Using extensive numerical simulation results, the proposed algorithm improves the imaging accuracy compared to the traditional methods without dense connections. In addition, the generalization ability test is carried out, reflecting a better image reconstruction than existing deep learning methods.

Keywords: Electrical Resistance Tomography, graph convolution network, optimization, Laplacian matrix.

Industrial Application: A5