Méthode automatique de détermination des paramètres d'inspection ultrasonore pour la reconstruction TFM

Abstract

This thesis presents the development of a complementary tool for the Total Focusing Method (TFM) reconstruction, an increasingly used imaging algorithm in ultrasonic non-destructive testing. Despite its ability to provide high-resolution images, TFM is highly sensitive to the measurement uncertainties of inspection parameters, requiring precise knowledge of propagation velocities and geometric parameters. The study focuses on defect visualization in a specimen with unknown thickness and propagation velocities using a probe positioned on an angle wedge. The challenge is to develop a fast, efficient, and automated method capable of determining these parameters from the Full Matrix Capture (FMC) data used for TFM image reconstruction. Two approaches were proposed to tackle this challenge. The first one, an analytical method, relies on comparing real and theoretical times of flight of structural echoes. However, it proves to be highly sensitive to experimental conditions, particularly the presence of coupling between the different components. Additionally, its computation time does not allow real-time optimization. The second method, based on a Convolutional Neural Network (CNN), demonstrates superior performance on real data, with fast computation and an uncertainty below 3.5%, enabling a highly accurate visualization of defects. These two methods are meant to be used complementarily. The first one, effective when the thickness is known, enables the creation of an experimental database with precisely known parameters. These data are then utilized to train a neural network capable of determining the parameters of a specimen with unknown thickness.

Date	4 Mar 2024
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Pierre Bélanger (Supervisor)