3G-1 Advanced Contour Detection by Locally adapted Dynamic Programming for Ultrasonic NDE Applications

During a continuous hydroforming process, a sheet metal contour is shaped. This highly reflecting contour shall be automatically tracked by means of ultrasound. The active fluid medium is a water-oil-emulsion at a pressure of several hundreds of bar. Following some specifications and limitations of the hydroforming procedure, the contour tracking system will consist of a low number of wide-angle ultrasonic transducers in an irregularly filled (quasi-sparse) array. The test geometries are hydroformed to specific, discrete stages. Thus, contour tracking experiments can be performed offline without the need of high pressure resistant equipment. The contour detection task is performed in two steps. Firstly, B-mode images (two- or three dimensional) are computed from a set of ultrasonic data applying the method of synthetic aperture focusing (SAFT). Separate transmitting and receiving transducers, mechanically moved to different aperture positions, are used. For each pair of transducer positions, pulse-echo data is recorded and stored for later image reconstruction. The respective experiments are performed in an experimental set-up with two 2 MHz transducers and two independent traversing units to generate an aperture. Secondly, the contour is automatically found in the computed images using an active contour model. This so called dynamic imaging approach is used here with respect to a priori information (e.g. curvature of the expected contour, shoulder of the contour, starting and final positions of the sheet metal, forming velocity, and preceding instantaneous contour). The algorithm is adapted locally, i.e. the curvature and smoothness is predicted for each point in the contour in contrast to global prediction. As contour tracking is performed over several forming states, the "history" of the contour is also considered. We will present the results of a contour tracking system for automated detecting and tracking of sheet metal surface contours during the discretized f- orming process (11 forming stages) by means of an active contour model. All presented results are based on measured ultrasonic data and will be discussed with respect to reconstruction accuracy, hardware requirements and computing time