Data-driven reconstruction method for electrical capacitance tomography

Abstract

The appealing superiorities, including high-speed data acquisition, nonintrusive measurement, low cost, high safety and visual presentation, lead to the success of the electrical capacitance tomography (ECT) technique in the monitoring of industrial processes. High-accuracy tomographic images play a crucial role in the reliability of the ECT measurement results, which provide the powerful scientific evidences for investigating the complicated mechanisms behind the behaviors of the imaging objects (IOs). Beyond the existing numerical algorithms that are developed for the solution of the inverse problem in the ECT area, a data-driven two-stage reconstruction method is proposed to improve the reconstruction quality (RQ) in this paper. At the first stage, i.e., the learning stage, the regularized extreme learning machine (RELM) model solved by the split Bregman technique is developed to extract the mapping between the tomographic images reconstructed by the some algorithm and the true images according to a set of training samples. At the second stage, i.e., the prediction stage, a new IO is reconstructed by the same algorithm used in computing training samples, and then the imaging result is considered as an input of the trained RELM model to predict the final result. The performances of the proposed reconstruction method are compared and evaluated by the means of the numerical simulation approach using the clean and noisy capacitance data with different noise levels (NLs). Quantitative and qualitative comparison results validate the practicability and effectiveness of the proposed data-driven reconstruction method. Research findings provide a new insight for the improvement of the reconstruction accuracy and robustness in the ECT area.