·Table of Contents
·Aeronautics and Aerospace
Real Time Digital Tomosynthesis System Dedicated to Industrial Solid Rocket Motor Examination
M. Antonakios, Ph. Rizo
LETI - CEA - Technologies Avancées 17, rue des martyrs F 38054 Grenoble cedex 9 France
SNPE Propulsion BP 57 33160 Saint Médard en Jalles Cedex
Usually, numerical radiographs of large objects such as solid rocket motors, show a poor contrast and a low signal to noise ratio images in the strongly attenuated area. Beside that, in the case of rocket motors, the totality of the object must be examined. The examination time must be as short as possible in order to improve the production rate. To reduce the noise on the images the images, it's necessary to sum static images or to apply imaging methods on moving object. To speed up the processing time of these tasks it's necessary to optimise the algorithms and to execute them on dedicated machines.
In this paper we present an original algorithm which implement a tomosynthesis method. The tomosynthesis allows to improve the quality of the radioscopic image and to provide three dimensional information. We propose herein a software and hardware architecture using this technique in its industrial development phase. This DTS (Digital Tomosynthesis System) allows a real time examination of moving objects.
This communication describes the industrialisation phase of a Digital Tomosynthesis System (DTS) which improves the quality of digital radiographs in the examination of Ariane 5 propellant motor. Each 3m diameter segment of the solid motor represents a volume of 100 m3 and a surface to be controlled of 100 m2. The system must detect defects with 2% contrast and save time on the inspection of these large objects.
To reach this goal we must apply the examination in real time during the motor rotation, but in this case classical methods of contrast and signal to noise ratio improvement are not able to satisfy the contrast constraint.
In order to combine the information between successive radiographs of each part of the moving object, a tomosynthesis method has been adapted to radial and tangential acquisition geometry. This method focuses on a given surface of the object and provides by reconstruction a sharp image of the structure on the scanned surface superimposed to a blurred image of the surrounding structures. Several hundreds of frames can be combined to compute tomosynthesis images. The signal to noise ratio and contrast detection obtained on processed images are equivalent to those obtained by averaging the same number of images with a static object.
In order to speed up the tomosynthesis reconstruction and to reach the real time examination during the object rotation, we integrated a DTS on a DSP based hardware, in a PC host. The geometric calibration parameters and pre-computed values are carried out off-line on the PC and loaded in the DTS. The on-line tomosynthesis calculation is produced by the multi DSP architecture which manage in real time, frame acquisition, parallel tomosynthesis calculation and output image display. This original and dedicated software and hardware architecture are integrated in an image computing system that control the tomosynthesis reconstruction and image monitoring.
In this paper, we present this DTS, its technical characteristics and performances, and the experimental images obtained on real motor examination in the industrial production site of Guyana.
2. Industrial control requirements
The ARIANE V solid motors are made of 3 meters diameters segments. Each 3m diameter segment contains 110 tons of solid propellant in a cylindrical metal envelope 10 mm on which is laid a thermal protection. A central channel allows the ignition and the combustion gas draining.
The control requirements are the detection of :
- Air cavities in the propellant mass. They must be detected if the defect size is larger than 20 mm. The total volume to be controlled for each segment is 100 m3.
- 1 mm delamination between the metallic structure and a thermal protection. The total surface to be controlled for each segment is 100 m2.
The general requirements are to control 100% of the booster and to immobilize the X-Ray area during 3 days at the most for a complete segment control.
It is therefore necessary to control a moving booster in front of the radiographic system. This real time control requires numerical radiographs and imaging methods to enhance the quality of the displayed radiographic images.
3. Control method
3.1 Tomosynthesis principle
The tomosynthesis technique, introduced in 1932 by Ziedses des Plantes , was used for laminography and longitudinal tomography. The principle is to accumulate the radiograph pixel values corresponding to the projection of the same point of the surface to be reconstructed. The capability to digitise radiographs, allows to generalise this approach to non parallel and non linear geometries .
We study the case of a cylindrical object moving around a rotation axis which is not in the projected area. The focal surfaces on which we apply the process are the radial planes containing the rotation axis. For each point of the observed plane, we follow its projection on the detector during the object rotation and sum in the focal plane all the contributions of this point. The projection trajectory of each point of the surface on the detector is determined off-line by a calibration measurement. In this way we correlate the information contained in the projections which contain the defect .
3.2 Tangential Tomosynthesis
The tomosynthesis principle presented previously is used for two different acquisition geometry. In the case of tangential tomosynthesis we search for delaminations between the metallic structure and the peripheral thermal protection. The rotation axis of the booster is outside of the field of view, and the projection image is the radiograph of a tangential area of the cylindrical object (fig 1). The general principle is to follow an observed plane during the rotation, and to compute an image of this plane projection, in the focal plane which is orthogonal to the X-ray source/detector axis.
Fig 1: Tomosynthesis tangential geometry|
3.3 Radial Tomosynthesis
The basic principle is the same as tangential tomosynthesis, but in this case the detector is located inside the booster in order to examine its mass (fig 2). Like in tangential examination, the movement of a radial plane of the object is followed during the rotation, and a projection image of this plane along the X-ray source/detector axis is computed.
Fig 2: Tomosynthesis radial geometry|
3.4 Tomosynthesis algorithm
The algorithm of tomosynthesis is based on a set of look up tables computed from the geometric parameters of the system, and a set of reconstruction parameters like the area of interest size, the number of position to be correlated. For each radiograph of an observed plane and for each angular position, we compute a look up table which will affect each point of the projection of this plan to the focal position. To obtain the tomosynthesis of the observed plane, the value of the designed pixel is simply summed to the corresponding value of the other angular position.
4. TOTEM : a Digital Tomosynthesis System
The real time system TOTEM has been designed once the tomosynthesis algorithm validation in deferred time has been shown. The detection system is the same as described previously. It produces the same kind of video signal. The dedicated system, controlled by a PC host, digitises computes and displays a processed image at the video rate. All the images are digitised and the refresh ratio of the result image is, according to the examination type (tangential or radial) from 4 to 10 images per second.
4.1 Hardware architecture
The system is designed around a mother board which contains the usual interface for a PC host connection on the PCI bus and 4 free slots for other connection modules. We use the capability of this board by connecting (fig 3) a video module for the image management and 2 process modules dedicated to the specific tomosynthesis process.
Fig 3: Hardware configuration of the industrial prototype|
4.2 Software architecture
The tomosynthesis examination is performed in three different steps. Calibration and look up tables computation are performed in deferred time (off-line), while tomosynthesis is performed in real time (on-line). We developed three different softwares (fig 4): the calibration software, the off-line software and the on-line software.
Fig 4: Software architecture for tomosynthesis examination|
4.3 On-line Processing parallelism
In the image processing and real time vision application field, the parallelism is referred to a limited number of schemes. Each one allows one or more efficient implementation. A set of these schemes has been formalised by Cole  and some models  (skeletons with synchronisation and communication attributes) have been proposed to ease the rapid prototyping and the application programming.
In our case if we consider the main loops of the tomosynthesis algorithm, we see that there is no complex operation. In fact, the characteristics of this algorithm is to perform all the calculation part in the off-line phase and to keep for the real time tomosynthesis a correlation with look up tables and a sum of images areas. Then this method fits with the split-compute-merge implementation model. This scheme (fig 5) is only usable when each data partition can be processed independently. The computing time on each area must be the same to maintain the harmony of the processor set and then to ensure the maximum efficiency.
Fig 5: Scheme split - compute - merge.|
Therefore, the images decomposition is handled in similar size data blocks, shared in the n processors memory space. Each processor executes the same process and returns its processed data block to the result fusion and image production dedicated processor.
4.4 Industrial validation
The industrial prototype of real time tomosynthesis will be installed in the industrial production site of Guyana (fig 6). The test motor segment contains some critical and calibrated defects. The contrast and signal to noise ratio improvement on these test defects has been verified in comparison with raw radiographs and sliding averaging images.
Fig 6: The industrial site of Guyanna|
Tomosynthesis software is launched at the start of an object examination. It computes the input image data flow of an analogic CCIR signal at the video rate (25 images per second) and produce an analogic PAL or SVGA format result data flow at the video frequency. The output rate is also 25 images per second and the refresh ratio change from 0.4 to 0.2 according to the tomosynthesis examination type. The size of the processed image is, in all the cases, 512 x 512 x 1byte. The displayed image have the size of the area of interest on which we apply the tomosynthesis. Typically the output image size is 250 x 400 for tangential tomosynthesis and 450 x 450 for the radial tomosynthesis.
We present here after some results carried out with the motor in tangential and radial position.
In tangential mode the defect is a 0.3mm delamination situated on the external side of the motor between the propellant and the metallic envelop (fig 7). On the left hand side view, the raw radiograph, and on right hand side a 36 planes tomosynthesis image. We clearly notice the enhancement of the signal to noise ratio and contrast, that makes easier the defect detection.
In radial mode, in order to simulate air cavities, we put in the propellant mass some polystyrene calibrated inserts (fig 8). On the right hand side view, the contrast improvement allows to detect a 10 mm cavity in 1 meter of propellant (density 1.7 g/cm3).
|Fig 7: Tangential examination and results |
|Fig 8: Radial examination and results |
The tomosynthesis principle leads to a sensible improvement of radiographic images signal to noise ratio and contrast. It allows the localisation and the measurement of observed defects inside the object. We applied this method for the examination of large moving objects such as solid rocket motors. The implementation in real time enhances the probability of detection in the same control time. The developed system can be used for the examination of other kind of moving objects. The integration of other algorithms, such as automatic defects detection, is a possibility for the future evolution.
This work has been partially granted by CNES.
- Ziedses des Plantes, "Eine neue method zur differenzierung in der roentgenographie", ActaRadiol., vol. 13 (1932) , pp 182-192.
- J. Liu, D. Nishimura, and A. Macovski, "Generalized tomosynthesis for focussing on an arbitrary surface", IEEE Trans. on Med. Ima., vol. 8, No. 2 (June 1989), pp. 168-172.
- Ph. Rizo, M. Antonakios, P. Lamarque, "Tangential radiography detectability by tomosynthesis techniques", QNDE 1994, Snowmass Co., USA, 1994
- M. Cole Algorithmic skeletons : structured management of parallel computation. Pitman/MIT Press, 1989
- Dunkan K.G Campbel. Toward a Classification of Algorithmic skeletons. Rapport de recherche YCS-276, Departement of Computer science, University of York,., Decembre 1996