NDT.net • May 2004 • Vol. 9 No.05
CT-IP 2003 Proceedings

Efficient Volume Digitizing with Adaptive Computerized Tomography

Alexander Flisch, Andreas Obrist, J├╝rgen Hofmann Swiss Federal Laboratories for Materials Testing and Research (EMPA) Section Electronics/Metrology, Ueberlandstrasse 129, CH-8600 Duebendorf, Switzerland Tel: +41-1-823 45 67, Fax: +41-1-823 45 79 e-mail: alexander.flisch@empa.ch
Corresponding Author Contact:
Email:

Abstract

Common industrial computed tomography (CT) scanners, which are suitable for big cast parts such as cylinder heads are usually equipped with a 450 kV X-ray tube and scintillator line detectors. The major drawback of these systems is that 3D data acquisition is a quite time consuming task, because an object has to be scanned slice by slice. Acquiring CT data from a whole cylinder head can take several hours if not even days.

In the context of the European research project FATIMA (First Article Tomography Inspection Methodology Advancement) EMPA developed software to accelerate the data acquisition procedure from line detector systems. Depending on the geometry of the cast part this so-called "adaptive scanning" permits a reduction of scanning time of about 10 to 50% with accuracy improved at the same time.

This contribution describes the method and shows applications from the casting of vehicle engines.

Introduction

Within the last five years industrial computed tomography (CT) evolved to a standard method for volume digitising. Its main applications are in automotive industries, where mostly (but not only) aluminium cast parts (e.g. cylinder heads) have to be inspected. Besides classic non-destructive examination (flaw detection) first article inspection (FAI) or wall-thickness analysis become of great interest. In FAI a real part is compared to its CAD model by means of its three-dimensional representation, commonly a point cloud. The result of this comparison is a colour-coded deviation map.

CT is the only non-destructive 3D digitising method that can easily reveal inner structures, too. Usually, CT scanners with line detectors and 450 kV X-ray tubes are used. These systems only provide enough penetration power needed for cylinder heads. Their disadvantage is (due to the line detector) that the data acquisition (and thus the digitalisation) has to be done slicewise. Therefore, it is slow and expensive. Digitising a whole cylinder head can take several hours if not even days.

For CT systems up to approximately 225 kV substantially faster flat panel detector systems are available today. Unfortunately, the penetration capability of these systems is too small for most applications in the field of vehicle engines.

The data acquisition can be done much faster without loss of accuracy if scanning is done with respect to the shape of the object. This can be done in three ways:

  1. Resolution optimised: It is not always necessary to scan the whole object with constant slice spacing. If geometry changes only little along the Z-axis, the slice distances can be increased and thus fewer slices have to be acquired. This, as a consequence, speeds up the scanning.
  2. Shape optimised: This is interesting only for translation scanning. The range of motion of the sample handler is extended or reduced according to the diameter of the part.
  3. Time optimised: In order to keep a constant signal to noise ratio, which is an expression for image quality, throughout the image stack, the integration time is adapted to the amount of material to be penetrated. Less absorbing material means shorter integration time and vice versa.
The software developed by EMPA uses resolution and/or shape optimisation in order to speed up data acquisition. Details can be found in the next section.

Adaptive scanning

EMPA developed an application, called SmartCT, which takes a CAD data file as input. The software then optimises the two scanning parameters slice spacing and range of motion (for translation scans only). Depending on the geometry of the cast part this so-called "adaptive scanning" permits a reduction of scanning time of about 10 to 50% with accuracy improved at the same time. Of course, the shape of the object must be known prior to scanning. But this is no problem in FAI as a CAD model must be available anyway.

Resolution optimised

In regions where the surface normals are almost perpendicular to the scanning plane the distance between the slices is reduced and thus the accuracy of the digitisation is increased. This optimisation is available for both, rotation and translation tomography.

Fig 1: Resolution optimised adaptive scanning: Few slices for the cylindrical part, many slices for edges.

Shape optimised

This optimisation method is interesting for scanning in translation mode only. If the diameter of the object exceeds the width of the detector line, rotation mode is impossible. Thus, the object has to be moved along the detector. The largest diameter determines the range of motion of the sample handler. If this diameter only covers a small fraction of the object it would be a waste of time to scan the rest of the object with the same range of motion. Shape adaptive scanning takes variations in diameter into account.

Integration time optimised

SmartCT does not yet support this method. Time adaptive scanning is useful not only to reduce scanning time but also to improve image quality. A longer integration time reduces the noise in an image. Longer integration may be necessary in regions where a lot of material has to be penetrated.

It should be well understood that the control software of a CT scanner must support adaptive setting of scan parameters.

Our SmartCT software package also includes a conversion application called SmartPoints, which generates a point cloud of the cast part surfaces based on the tomograms. Contrary to commercially available software today the new product can process a stacked tomogram sequence with arbitrary slice distances. For adaptive scanned tomogram stacks this conversion application is a mandatory addition. Besides this it has the following further advantages:

  • Data sets of any size can be processed off-line in a few minutes.
  • For each slice an optimal threshold value is determined prior to segmentation.
Therefore, a higher accuracy is obtained compared to applying a global threshold value to the entire stack.

Results

Our SmartCT method was successfully applied to different cast parts. As an example we want to present the results for an aluminium cylinder. Its dimensions are 180x150x130 mm (WxLxH).

Fig 2a: Aluminium cylinder 180x150x130 mm (object with courtesy of Bombardier-Rotax GmbH, Austria). Fig 2b: The optimal slice spacing as reported by the SmartCT software. The larger step size used to scan the cylindrical top can be recognised.

Fig. 2a shows a side view of the object. It consists of a cylindrical and a more complex part. The original CAD drawing was exported to STL format, which could be used as input for SmartCT.

After SmartCT processing the resulting slice spacing shown in Fig. 2b was obtained. The desired slice spacing was set to 0.4 mm. As you can see, the software suggests smaller step sizes in regions were edges occur. On the other hand, the cylindrical top where geometry changes only little is scanned with steps of 1.6 mm.

After the scan the image stack was converted to a point cloud. This was done with our SmartPoints application. This software uses the SmartCT parameter file for the calculation. As mentioned above, contrary to commercially available programs this new product can process a stacked tomogram sequence with arbitrary slice distances. Fig. 3 shows the obtained point cloud. A thinned out cloud (about 500,000 points) is shown for visualisation reasons. The full cloud comprises more than 1 million points.

Fig. 3 Point cloud generated with SmartPoints: The point density varies throughout the cloud. This is due to the adaptive slice spacing! The smaller the step size is the denser the point cloud gets!

The following table compares a scan with constant slice spacing of 0.4 mm to a scan with optimised parameters. The whole object could be scanned in a third of the time needed if no optimisation was made. This is of a great benefit for our customers because the costs are significantly reduced.

Slice spacing fixed 0.4 mm with SmartCT Saving
Number of slices 325 115 65%
Scan time [h] 13 4.5 65%
Costs [Euro] 2'800 1'400 50%
Table 1 Comparison of scan time and costs without and with SmartCT.

Discussion

SmartCT can significantly reduce the time needed to scan a complex cast part. Depending on the inner and outer geometries of the object a reduction of about 10 to 50% is feasible. The optimisation algorithm is based on CAD data. In critical regions (e.g. parallel to the scanning plane) a smaller step size refines data acquisition and improves accuracy. Partial volume effects are minimised, too.

The other application, called SmartPoints, generates a point cloud from image stacks scanned with variable slice spacing. In order to improve accuracy a threshold value can be applied to every single slice.

The benefits are:
1. Shorter data acquisition time
2. Reduced costs for the customer
3. Improved accuracy

Link

The work was carried out within an EC Framework 5 project called FATIMA (First Article Tomography Inspection Methodology Advancement). Further information can be found under www.eufatima.ch

© NDT.net |Top|