Non-destructive testing by supervised analysis of 3D ultrasound images.

Summary The purpose of this thesis is to develop a processing chain to extract useful information from 3d ultrasonic data and to characterize any defects present in the inspected part. This characterization was approached for cracks controlled by the same transmitter/receiver. In a first part, we recall the principles of ultrasonic non-destructive testing as well as the classical representations of ultrasonic data. The second part is devoted to the study of a model of extraction of the echo information present on the data by means of an adapted wavelet base. The use of a single wavelet translated in time is made possible by working on a complex representation of the original real data. A first step allows to detect and position the echoes of significant amplitude. In a second step, a spatially consistent regularization of the detection times is performed using a Markovian model. This eliminates echoes whose detection times are not part of regular time surfaces. The following sections deal with the localization and sizing of cracks. Features extracted from the ultrasonic beam are used to determine the path of the ultrasonic wave from the sensor to the diffracting object when the echo response is maximum. The time of detection obtained for this echo is matched with the time of travel along the defined path to position an edge point in the part. We thus obtain a set of discretization points for each edge. In the framework of 3d data obtained on an isotropic material, the extreme edge points are eliminated by using a comparison criterion on the echodynamic curves associated with the detection points on real data and on equivalent simulated data. Localization is discussed for cracks located in an isotropic material or anisotropic coated steel.