ECT sensor design using machine learning techniques

Within the framework of image reconstruction by Electrical Capacitance Tomography (ECT) sensing, we investigate the relevance of the sensor structure embodied in both the number and the size of the electrodes. While most of the studies in the literature exhibit sensors with an assumed arbitrary structure, we consider that these two properties possess a significant impact on the sensor performances. In our study, the emphasis is about the detection of a single bubble with random size and position. We propose to determine the architecture of the sensor that leads to the most accurate image reconstruction. To achieve this objective, we propose to determine the image from a set of independent measurements using LS-SVM models selected with a sophisticated validation method. Indeed, this way to proceed ensures a faster image computation than inverting an underdetermined set of linear equations. By doing so, the computational burden is reduced since it leads to a direct calculation instead of an iterative optimization. Various numerical experiments are presented and discussed. They show the effectiveness of our assumption.