In this work a novel approach to solve the 2D Electric Capacitance Tomography (ECT) inverse problem is presented. The aim of this work is to improve the spatial resolution of the ECT method to test olymeric components for large-scale manufacturing and assembling industry,preserving the time resolution and the deterministic computational time of this non-destructive technique. The data obtained from Finite Elements techniques have been used to train an Artificial Neural Network (ANN) in order to solve the direct problem, namely calculating the capacitance for a given permittivity of the medium affected by defects in different positions and dimensions. In particular, the dimensions and the barycentre positions of the defect are the input of the ANN and the simulated capacitances represent the output. Once the ANN has been trained, in order to perform the test of the material, an inverse problem has to be solved,specifically determining the characteristics of the defect on the basis of the measured capacitances. To do this, the trained ANN can be exploited, by inverting the neural network instead of the problem itself. Results showed good accuracy in the identification of the position of the defects.
A neural network based approach to solve the electrical capacitance tomography inverse problem
CARCANGIU, SARA;MONTISCI, AUGUSTO
2013-01-01
Abstract
In this work a novel approach to solve the 2D Electric Capacitance Tomography (ECT) inverse problem is presented. The aim of this work is to improve the spatial resolution of the ECT method to test olymeric components for large-scale manufacturing and assembling industry,preserving the time resolution and the deterministic computational time of this non-destructive technique. The data obtained from Finite Elements techniques have been used to train an Artificial Neural Network (ANN) in order to solve the direct problem, namely calculating the capacitance for a given permittivity of the medium affected by defects in different positions and dimensions. In particular, the dimensions and the barycentre positions of the defect are the input of the ANN and the simulated capacitances represent the output. Once the ANN has been trained, in order to perform the test of the material, an inverse problem has to be solved,specifically determining the characteristics of the defect on the basis of the measured capacitances. To do this, the trained ANN can be exploited, by inverting the neural network instead of the problem itself. Results showed good accuracy in the identification of the position of the defects.
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
