In order to reduce the photonic noise on the images and to improve the signal to noise ratio of the defects, a method of sliding averaging is usually applied. This method sums few digital radiographs acquired at different positions assuming that the displacement of the object between the different radiographs is negligible. The problem of this method is that the number of radiographs that can be summed is very small and therefore the signal to noise ratio improvement is not enough. Most of the time with this method, no more than 6 radiographs can be summed, the signal to noise ratio is thus improved by 2.5.To overcome this limit, we developed a dedicated digital tomosynthesis processing of the radiographs.

First we recall briefly the principle and the limits of the tomosynthesis which combine a large number of radiograph in order to increase the signal to noise ratio of the defects. The number of radiographs that can be combined is usually larger than 100. The possible signal to noise ratio improvement is a little smaller than the square root of the number of projection, because the attenuation of the X-ray is not the same in each direction. We show that there is an optimum number of projection for a given acquisition geometry.

Second we present our experimental setup and the results obtained on a set of data acquired on a solid rocket motor phantom of 3 metres in diameter with a 15 MeV linear accelerator source and a high energy digital radiographic system based on a Gd₂O₂S: Tb converter screen and a high sensitivity silicon target intensified camera. The method has been implemented on a work- station. A specific geometric calibration procedure has been developed. On tomosynthesis images, we clearly showed an improvement of the defect detectability and of the signal to noise ratio. On 1.5 mm delamination the signal to noise ratio is improved from 1.5 for the radiograph to 8 for the tomosynthesis image. Holes of 16 mm in diameter not detected on the radiographs or on the sliding averaging images are clearly seen on the tomosynthesis image.

Finally we examine the potential new application for this technique to other industrial fields such as the welding monitoring. We will also examine the possibility to develop a dedicated hardware which would perform the tomosynthesis processing in real time.