IOP PUBLISHING MEASUREMENT SCIENCE AND TECHNOLOGY

Meas. Sci. Technol. 19 (2008) 094002 (11pp) doi:10.1088/0957-0233/19/9/094002

An ultra fast electron beam x-raytomography scannerF Fischer1, D Hoppe1, E Schleicher1, G Mattausch2, H Flaske2, R Bartel2

and U Hampel1

1 Institute of Safety Research, Forschungszentrum Dresden-Rossendorf e.V., PO Box 51 01 19,01314 Dresden, Germany2 Fraunhofer Institute of Electron Beam and Plasma Technology (FEP), Winterbergstraße 28,01277 Dresden, Germany

E-mail: F.Fischer@fzd.de

Received 21 January 2008, in final form 14 April 2008Published 24 July 2008Online at stacks.iop.org/MST/19/094002

AbstractThis paper introduces the design of an ultra fast x-ray tomography scanner based on electronbeam technology. The scanner has been developed for two-phase flow studies where framerates of 1 kHz and higher are required. Its functional principle is similar to that of the electronbeam x-ray CT scanners used in cardiac imaging. Thus, the scanner comprises an electronbeam generator with a fast beam deflection unit, a semicircular x-ray production target madeof tungsten alloy and a circular x-ray detector consisting of 240 CZT elements with 1.5 mm ×1.5 mm × 1.5 mm size each. The design is optimized with respect to ultra fast imaging ofsmaller flow vessels, such as pipes or laboratory-scale chemical reactors. In that way, thescanner is capable of scanning flow cross-sections at a speed of a few thousand frames persecond which is sufficient to capture flows of a few meters per second velocity.

Keywords: electron beam tomography, x-ray tomography, multiphase flow measurement

(Some figures in this article are in colour only in the electronic version)

1. Introduction

In many engineering fields, such as chemical and processengineering, nuclear engineering and mineral oil production,there is strong interest in visualizing and assessing the innerstructure of multiphase flows and distributions of liquids, gasesor particle suspensions. Examples for multiphase flows are,for instance, steam–water or gas–water flows in pipes, bubblecolumns, valves, pumps or heat exchangers but also dynamicmixtures in reactors, rectification columns and other chemicalprocessing equipment. Proper knowledge of flow behaviourand flow patterns under certain process conditions is not onlyessential for optimization, monitoring and assessment of safetyissues, but also scale-up issues related to laboratory equipment.Therefore, dedicated measurement techniques are required togain a closer look inside such processes, thereby providingdata for improved modelling and CFD code validation.

X-ray CT as an imaging modality is highly advantageousdue to its non-intrusiveness and its ability to penetrate opaquewall materials. One essential disadvantage of existing CT

systems is the requirement for rotating components. Tomeasure multiphase flows in a velocity range of one meterper second or more, frame rates of at least 1000 framesper second are required to produce sharp phase distributionimages with a spatial resolution of about one millimetre. Toachieve this, mechanically rotating parts are to be avoided.Scanned electron beam x-ray tomography is therefore apromising technology. Instead of mechanical rotation ofscanner components, an electron beam is rapidly swept acrossan x-ray target using deflection coils. This technology wasintroduced in medicine more than two decades ago whereit is mainly being used for cardiovascular diagnostics [1, 2].However, medical systems with frame rates up to 20 frames persecond (in the following: fps) are still too slow for technicalflow diagnostic problems. There are a few other notableapproaches to fast tomography using x-rays or gamma rays.Johansen et al [3] introduced an isotopic CT scanner with fivestationary Am-241 sources delivering up to 100 fps. This hightemporal resolution is achieved at the cost of reduced spatialresolution since only five projections are acquired. Morton

0957-0233/08/094002+11$30.00 1 © 2008 IOP Publishing Ltd Printed in the UK

Meas. Sci. Technol. 19 (2008) 094002 F Fischer et al

et al [4], Misawa et al [5] and Hori et al [6] introduced multiplex-ray tube scanners operating with multiple gated x-ray tubeswhich are switched sequentially to produce the tomographicdata. Hori et al [6] claimed to reach a frame rate of2000 fps. Another approach was introduced by Hampel et al[7–9] who used a linearly scanned electron beam to producelimited view data of a dynamic object at very high speeds ofup to 10 000 fps.

Since all of the above-mentioned approaches havenumerous drawbacks and limitations regarding multiphaseflow tomography, it was decided to design and build up adedicated electron beam x-ray CT scanner for this application.The decision for an electron beam type CT and against a CTwith stationary gated sources was due to several reasons. First,electron beam technology provides excellent versatility sincethe electron beam can be almost arbitrarily steered and shapedas needed. In the simplest case the electron beam will beswept along a semicircular target as in medical electron beamCT to generate projection data from a single plane. Moreover,subsequent sweeping on different circular paths on the targetenables multi-slice CT as well as velocity measurement fortwo-phase flows. Additional beam optics can be providedto shape the focal spot as required and to adjust the beamfocus when scanning at different axial positions (planes).Only one beam power supply is required. Another strongargument is the straightforward upgrade capability towardshigher x-ray energy by introduction of high-energy electronbeam generators or superconducting accelerators.

This paper introduces the developed scanner primarilywith respect to its mechanical design, its electrical componentsand its imaging performance. Therefore, the next threechapters introduce the general system design, the electronbeam generator and the x-ray detector. This description isfollowed by explanation of the scanning procedure and the dataprocessing. Eventually, we will discuss the most importantperformance parameters and show examples of scanned andreconstructed dynamic objects.

2. Scanner principle and the basic system design

Figure 1 shows the operation principle of the ultra fast scannedelectron beam CT. An electron beam of sufficient energy isproduced by an electron gun, focussed onto a semicircular x-ray production target which surrounds an object of interest.The electron beam is swept rapidly across this target by meansof an electromagnetic deflection system and in this way x-rays are generated from a moving focal spot. Radiation thatpasses the object is recorded by a fast x-ray detector. Thedetector is designed as a circular ring, mounted inside thescanner head with some axial offset relative to the focal spotpath. In contrast to medical electron beam CT, the setup isoptimized with respect to ultra fast imaging, which implies asmaller source–detector separation and shorter electron path,accordingly. The setup can be used, as shown in the sketch, tovisualize multiphase flows in pipelines or vessels.

The design of our ultra fast electron beam CT scanneris illustrated in figures 2–4. Figure 2 shows an explodedview CAD drawing, figure 3 shows a block scheme with the

Figure 1. Operation principle of ultra fast electron beam x-ray CT.

essential scanner components and figure 4 shows photographyof the scanner mounted on a vertical pipe. The wholetomography system consists of the scanner itself, a high-voltage generator, the operating rack and a measurement PC.The high-voltage generator provides the electron accelerationvoltage of up to −150 kV at a maximum beam current of65 mA (10 kW maximum electron beam power) along withall necessary auxiliary supplies, such as Wehnelt voltage,bombardment voltage and filament heating current. Thecontrol rack comprises coil current amplifiers, vacuum pumpcontroller, beam monitoring devices, and a control PC. Bothhigh-voltage generator and control rack with its componentsare more or less commercial products. The scanner can besubdivided into three main parts: the electron beam gunincluding beam optics, the beam tube and the scanner head withthe target, the detector and the beam monitoring equipment.These components will be described in more detail furtherbelow. The diameter of the scanner at the beam tube is273 mm and the overall length is about 1500 mm. The freeelectron travel path, i.e. the distance between anode and target,is about 1000 mm.

3. Electron beam scanner and target

The electron beam gun is a triode system operated witha small-size tungsten bolt cathode that is indirectly heatedby electron bombardment. The Wehnelt electrode and theanode are shaped as a Rogowski transducer with a bell-shapedthrough-hole anode. This configuration has been chosen since

2

Meas. Sci. Technol. 19 (2008) 094002 F Fischer et al

Figure 2. CAD drawing of the scanner (exploded view): (1) electron gun with beam optics and vacuum pump, (2) beam tube, (3) targetholder with target and cooling circuit, (4) x-ray detector.

Figure 3. Block scheme of scanned electron beam x-ray CT scanner.

it greatly reduces the dependence of focal spot size on electronbeam current. The electron beam gun is operated at about10−6 mbar gas pressure which is provided by a vacuum pumpsystem, consisting of a scroll pre-pump and a turbo molecularmain pump operated in series. The gas pressure in the beamcolumn and the scanner head is somewhat higher in the range of10−5 mbar.

After passing the anode hole, the accelerated electronspass the beam forming section which contains centring,focussing and deflection coils. The quadrupole centring coilis used to align the beam exactly with the optical axis ofthe focussing and deflection system. This is necessary tocompensate the effects of slight geometric misalignmentsin the electron optics which would cause beam astigmatismand aberration. Once centred, the beam is focussed by thefocussing coil package to a small focal spot on the target.The scanner currently achieves a spot size of about 1 mmwhich is determined by the scanner design. Maximum beamdeflection angle is ±17.5◦ in both x and y directions. Thecut-off frequency of the deflection system is 10 kHz. Thefar end of the electron beam column has been designed in ahorse shoe shape with an opening that enables the scanner toeasily accommodate pipes, vessels, or other objects. Inside thescanner head resides the x-ray target, which is a semicircular

metal ring of 240◦ opening angle and 256 mm outer diameter(figure 5). The target has an inclination of 60◦ relative to thebeam axis. As can be seen in the schematic drawing in figure 1,this arrangement allows scanning of axially extended objects.In turn, it requires an elliptical beam deflection pattern toproduce a circular focal spot path on the target. The target is acompound of a massive copper body with tungsten alloy platesat the top surface. Such a combination is typical for heavy dutyx-ray targets, where the tungsten with its high melting pointenables highest dissipation of beam energy, whereas the copperas a superior heat conductor allows efficient heat removal. Toremove the heat, the target is cooled by means of a watercooling circuit. In the measurement plane, an x-ray window isprovided by appropriate reduction of the thickness of the innerscanner head wall to 0.7 mm. The x-ray detector is arrangedinside the opening close to the inner scanner head wall at adiameter of 132 mm. This gives a free space of about 120 mmdiameter which defines the maximum diameter of objects.

4. X-ray detector

The x-ray detector resides inside the horse shoe enclosureand is made separable to ease dismantling and mounting ofthe scanner head. The detector ring comprises 240 CZT

3

Meas. Sci. Technol. 19 (2008) 094002 F Fischer et al

Figure 4. Photograph of the scanner at a vertical pipe.

Figure 5. X-ray target mounted inside the scanner head.

4

Meas. Sci. Technol. 19 (2008) 094002 F Fischer et al

Figure 6. Block scheme of the fast multi-channel x-ray detector electronics.

pixels, each of 1.5 mm × 1.5 mm × 1.5 mm size. Thedetector is arranged at a slight axial offset of 5 mm relativeto the plane of the focal spot path. The CZT crystals areoperating in current mode. A coarse block diagram of signalprocessing hardware is given in figure 6. Each detector pixelis connected to a gain-selectable current-to-voltage converter.Each converter consists of three stages: a transimpedanceamplifier, followed by a gain-selectable amplifier, followed bya buffer amplifier. The transimpedance amplifier transformsthe electrical current generated by the x-rays in the CZT pixelinto a voltage. The gain-selectable amplifier provides a secondamplification stage which is selectable with respect to gain andbandwidth in steps of 5/500 kHz, 5/50 kHz, 25/5 kHz and50/500 Hz. This allows us to choose an optimum operatingspeed and integration time constant for the given imagingproblem. The preamplifier is followed by a post-amplifierwith 500 kHz bandwidth. The amplified signal is convertedby a 12 bit analogue-to-digital converter (ADC) and fed intoa temporary data RAM. This system consists of 512 parallelchannels. Amplifiers, ADCs and RAM modules are controlledby a single microcontroller and the communication betweenthe PC and the detector electronics as well as the transfer ofmeasurement data is done via the USB 2.0 interface protocol.The maximum sampling rate of the detectors is 1 MSample/s.The electronics, including the ADCs, is placed close to thedetector ring at the head of the scanner, whereas the digital partof the electronics (multiplexer, RAM modules, microcontroller

and USB 2.0 interface) along with the power supply unit, islocated some distance away in a special 19′′ rack.

5. Additional features

Especially for the tuning of the electron beam, additionalbeam monitoring components are required. For that reason, asmall camera with illumination is mounted inside the vacuumenclosure. It enables visualization of the focal spot path byvirtue of the visible light emitted from the focal spot. Since thiscamera cannot provide sufficient information on the focal spotsize and shape, two semicircular electron catchers surround thetarget. These are sheets of molybdenum which are groundedvia shunt resistors. The electron charge collected by thecatcher produces a measurable voltage drop across the shuntwhen it is drained to ground. In this way, it is possible to obtainthe beam profile by scanning across one of the edges. On theother hand, one can get some helpful information during theprocess of manual focal spot path tuning. A planned featureis that of using backscattered electrons to monitor the electronto x-ray conversion in the focal spot which may change due todegradation of the target surface.

Another feature is a protective slit mask with beam dumpcapability mounted halfway between the anode and the target.It is made from a massive copper body and comprises a slitwhich tightly restricts the electron beam to the target surfacearea. It helps to avoid thermal destruction of the scanner’s hullin case of a malfunction of the beam steering system or an

5

Meas. Sci. Technol. 19 (2008) 094002 F Fischer et al

accidentally wrong manual steering operation. Furthermore,this mask can be used as a beam dump to ‘park’ the electronbeam at full power. Therefore, the dump is connected with thewater cooling circuit used for target cooling.

All scanner parts are controlled from the 19′′ rack whichcontains the control components, such as the coil currentamplifiers, the vacuum pump controller and the control PC.The measurement and control software is implemented inLabVIEW, except for the detector data acquisition and imagereconstruction software, which is implemented in C++.

6. Scanner operation and data processing

In the scanning mode, the electron beam is swept across thetarget in a circular way. When passing the open part of thetarget, the beam is properly masked out by the slit mask (beamdump). The beam deflection figure is an ellipse, given by

αx(t) = x0 + a cos ωt αy(t) = b sin ωt. (1)

Here, αx and αy denote the deflection angles produced bythe x- and y-deflection coils, t is the time, and ω is the angularfrequency. The ellipse has a constant offset x0 in one directionsince the optical axis (axis of undeflected electron beam) doesnot coincide with the target ring centre. a and b define theelongations of the ellipse. The three latter parameters aredetermined by geometrical relations, i.e. from the target ringdiameter, the distance between the focus of the deflection coneand the target and the target plane inclination relative to theoptical axis of the scanner. Note here that the focal spot pathgeometrically represents a cut through a skewed elliptical cone,which makes its mathematical description quite complex. Thescanning pattern, once fine-tuned, is stored in the control PC asa sequence of tuples {αx,i , αy,i}. During a scan, this sequenceis repeatedly delivered as input voltages to the coil currentdrivers via a two-channel DAC board at a user defined rate.

A tomographic scan is essentially performed bysimultaneous switching of the Wehnelt voltage, deflectionpattern generator and triggering of the detector dataacquisition. Practically, however, it takes around 1 s tostabilize the electron beam after it has been turned on. Usually,the detector is therefore triggered with a correspondingdelay. For the sake of simplicity, the deflection signal isnot synchronized with the detector data acquisition directlybut instead on two extra channels of the detector electronicsthe deflection voltage signal produced by the deflection coilcurrent amplifiers is acquired. With the given RAM capacitya scan as long as 10 s at full 1 MHz sampling rate is possible.After each scan, the data RAM are read out via the USB2.0 interface and the entire data set can be reconstructed.

The data processing is only schematically described, sinceit is quite similar to that of other CT systems. Once a scan hasbeen performed and data have been transferred to the PC, thefollowing processing steps need to be executed to reconstructan image. First step: the continuous data sequence is brokeninto sets of size ND × NT where ND denotes the number ofdetectors and NT the number of equidistant temporal pointsfor one complete electron beam revolution. This is achievedby evaluation of the deflection signal recorded with the two

Figure 7. Geometry of the scanning plane.

extra detectors. Each of those data sets then represents theraw data for one frame of the image sequence. Second step:The raw data of each frame are mapped from the temporaldomain into the angular domain of the target. That is, onecalculates a projection data matrix of size ND × NP where NP

is the number of equidistantly distributed source positions onthe target. Note that this transformation is a comparativelycomplex nonlinear one which is necessary due to the fact thatthe source trajectory on the target is a cut of a skewed ellipticalcone. Third step: the data can now easily be resorted into a fanbeam data set as for a conventional CT scanner. Thereby, alldata from source positions outside the valid angular scanningrange of 240◦ are discarded. Fourth step: for the fan beamdata of each frame the line integral x-ray attenuation valuesare calculated according to

Em,n = − logIm,n − I (d)

m

I(0)m,n − I

(d)m

. (2)

Here, I denotes the x-ray intensity (which is represented bythe respective acquired amplifier output voltage), m and n arethe indices of detector and projection, superscript (d) denotesa previously acquired dark reference and (0) a previouslyacquired reference measurement with no object in the scannedplane. The last step is image reconstruction. The geometryunderlying this procedure is schematically depicted in figure 7.From the focal spot of the electron beam, the shadowgraph isprojected onto the detector ring via the radiation fan indicatedin the figure. Therefore, with the given data, one has to solvea fan beam reconstruction problem. To apply the standardconvolution-backprojection technique, the fan-beam-type dataare mapped into the Radon space of virtual parallel beamprojection geometry. This virtual projection space is sampledat Nθ projection angles equally distributed across the rangefrom 0◦ to 180◦ and NS values of axial beam offset. The latter is

6

Meas. Sci. Technol. 19 (2008) 094002 F Fischer et al

determined from the relation NS = Dsupp/�S, with Dsupp beingthe diameter of the support (defined by the extension of thescanned object in the scanning plane) and �S the ray spacingwithin a projection. �S is user-selectable and is commonlyin the order of 0.5 mm which corresponds to the achievablein-plane resolution. The number of projections is then set toa rounded value of Nθ = πNS which gives a number of NS

sampling points along the circumference of the support. Thespatial extension and pixel number of the reconstructed imageare again user-selectable. For the results presented later, animage size of 80 mm × 80 mm with 128 × 128 pixels waschosen.

The fan-beam-type data were mapped into virtual parallelprojection space providing a mapping operator M whichtransforms the data from the fan beam extinction matrix E(fan)

into a parallel beam sinogram matrix E(par) according to

E(par)k,l =

NS−1∑

n=0

ND−1∑

m=0

Mk,l,m,nE(fan)m,n . (3)

The tensor M is sparse and the weights Mk,l,m,n aredetermined once before reconstruction from geometricalconsiderations. Thus, given a ray in virtual parallel projectionspace, the four closest corresponding rays in fan beamspace were sought to determine the weights according to thedistances of these rays from the given ray. The closest raysand the distances are found by considering the intersectionpoints of the parallel beam ray with the detector circle and thesource path and by finding the corresponding neighbouringsource positions and detectors around the intersection points,respectively. Note that some parallel rays will havetwo corresponding fan beams due to some source–detectorsymmetry. In this case, averaging is performed.

The transformed data are eventually reconstructed bya convolution-backprojection algorithm. First, the data areconvolved row-wise according to

∀ k : E(par,filt)k,l =

NS−1∑

p=0

H(l+p) mod NSE

(par)k,(l+p) mod NS

. (4)

Here, H is the reconstructing kernel which is the inverseFourier series of a frequency ramp modified by some apodisingwindow. For the results presented later, the Hanning windowwith a cut-off frequency of approximately 0.5 line pairs permillimetre (in the following: lp mm−1) is used. The last stepis then the backprojection of the filtered data into the imagearray according to

µi,j =Nθ−1∑

k=0

NS−1∑

l=0

Bi,j,k,lE(par,filt)k,l . (5)

B is the backprojection operator and its values are definedby the shares of each pixel (i, j) with ray (k, l) and µ isthe reconstructed image. In principle, the above-describedalgorithm is the standard algorithm of transmission CT andcan be found as such in many textbooks, such as [10].

Figure 8. Time response of a representative CZT detector element.The rise time constant is τ = 450 ns.

7. Performance tests

7.1. Detector time response

The scanner has been designed for a maximum detectorsampling rate of 1 MHz and a corresponding temporalbandwidth of 500 kHz. This means that at a scan rate of5 kHz there will be 200 projections available forreconstruction, which is sufficient for a spatial resolutionof about 1 mm. The CZT material is known to havedifferent electron and hole mobilities, which leads tothe question of whether its temporal response can matchthe detector bandwidth. To check this issue the step responseof the given CZT crystals was measured, which is easilydone by fast scanning of the electron beam across a slitmask mounted temporarily across the target. A maximumscanning frequency of 10 kHz (7 mm µs−1 focus velocity)was chosen to ensure prompt x-ray exposure on the detectorpixels. The measurement was performed with a fast amplifieroperated at 50 MHz bandwidth. Figure 8 shows the measuredtime response function of one representative detector channel.Signal rise time is about 1.8 µs and fall time 1.7 µs for theinterval 10% to 90%, giving a time constant of τ ≈ 450 ns.This proves that the pixel’s time response is sufficiently fastfor our detector system.

7.2. Focal spot size

The focal spot size of the electron beam on the target can bemeasured by analysing the electrical signal from the electroncatcher when the electron beam crosses an edge. Figure 9shows a representative edge signal. The size of the focal spotis related to the time constant of the edge response function by

ds = 2 · vB · τ√π

. (6)

7

Meas. Sci. Technol. 19 (2008) 094002 F Fischer et al

Figure 9. Edge response function of the focussed electron beamwhen passing an electron catcher edge.

Figure 10. Modulation transfer function obtained from areconstructed object edge.

Here, vB is the velocity of the focal spot and ds is theeffective spot diameter. With a velocity of vB = 424 m s−1

for this particular experiment, a beam diameter of 1.1 mm wasobtained. Note that the beam diameter has approximately thesize of the cathode bud, which is typical for this electron opticsystem. In principle the focal spot can be made smaller bya different electron optics design, which may be a topic forfuture work.

7.3. Spatial resolution

Spatial resolution within the reconstructed images has beendetermined from the edge spread function of a reconstructedrectangular object made of acrylic glass (figure 10). Therefore,the grey value profile along a horizontal line through one ofthe edges of the object was extracted. From the edge spread

Table 1. Mean and standard deviation of pixel values in figure 11.

Pixel Mean Standardnumber value deviation

1 1.3 0.492 10.4 0.343 21.8 0.374 32.8 0.455 40.2 0.43

function esf (x), derive the point spread function psf (x) viathe differentiation

psf (x) = ∂

∂xesf (x). (7)

The measured point spread function is Fourier-transformed to yield the modulation transfer function (MTF),as shown in figure 10. From the MTF, we get a value for thespatial resolution of the scanner of 0.51 lp mm−1 at 10% MTF.

7.4. Image noise

To assess the image noise a phantom, as shown infigure 11 (top), was used. It consists of four tubes eachwith an inner diameter of 8 mm, which were filled withsubstances of different attenuation coefficients. At the rightthe reconstructed slice image is shown, scanned at a rate of1 kHz. The tube denoted by 2 was filled with pure water,whereas tubes number 3 to 5 were filled with a mixture ofwater and iodine contrast agent (PERITRAST R© Dr FranzKohler Chemie GmbH). The volumetric concentration of thecontrast agent was 50% for tube number 5, 33% for tubenumber 4 and 16.5% for tube number 3. Number 1 denotesa background pixel monitored as a reference. Figure 11shows the time plot of the pixel values. The measuredattenuation ratios for tubes 2, 3 and 4 agree quite well with thegiven values. The value for tube 5 with the highest contrastagent concentration is slightly underestimated, which is to beattributed to yet uncorrected beam hardening. The purposeof this experiment was to assess the temporal deviation ofthe pixel attenuation values. The spread of the gray valuerelative to its mean is highlighted by the histograms at the rightof the plot. Quantitative analysis is given in table 1. Note: apractically relevant case is discrimination of water and gas, i.e.values (1) and (2). Here, we find that the standard deviation ofthe water attenuation value in pixel (2) is 3.3% relative to themean attenuation difference between water and air. This showsgood discrimination capability of the tomographic scanner.

8. Application examples

As an illustration, two application examples are introducedhere. The first is a dynamic particles phantom which has beenused in earlier studies with fast limited-angle-type tomography[7]. The phantom is made of a circular aluminium housing(cup) of 1.5 mm wall thickness and 40 mm outer diameterwhich contains three particles. These particles are set inmotion with a dc motor driven disc that rotates inside the

8

Meas. Sci. Technol. 19 (2008) 094002 F Fischer et al

Figure 11. Image noise assessment: In the top, the phantom with four tubes filled with four different substances along with a reconstructedslice image is shown. The bottom plot shows the values of the central pixel of each absorber and one background pixel in subsequent framesfor 1 kHz scan rate. The histograms at the right illustrate the spreading of the reconstructed attenuation coefficient.

cup. A centrally placed stirrer forces the particles to irregularmovements by rotation. As test particles, glass beads of4.5 mm diameter which possess through-holes of 1 mmdiameter were used. The phantom may serve as an exampleof solid particle movement in a vessel. A scan was performedat 1 kHz frame rate and 500 frames altogether. The speed ofthe stirrer was set to 20 rotations per second, which gives50 frames per stirrer revolution. Figure 12 shows someselected images of the whole sequence. All parts of thephantom can clearly be reconstructed and the movement of thepearls can be tracked in high detail. Except the limits of spatialresolution, all details are reconstructed with no noticeableartefacts. The processing of the data and the reconstructiontook about 5 min on a standard PC.

Another application example is a two-phase flow in abubble column vessel. The column is of 60 mm inner diameterand 500 mm height and made of Perspex. Gas was injectedat different flow rates from the bottom by a single injectorneedle, and the measurement was performed 150 mm abovethe bottom of the vessel. Figure 13 shows the reconstructedgas distribution as a central axial cut image for a moderategas volume rate with few larger bubbles. Again, the scan wasperformed at 1000 fps for the 0.5 s scan duration. As can beseen, all bubbles are well-resolved and even the thin liquidfilm between the single bubbles is reconstructed. Further,three-dimensional image processing allowed us to extract otherfeatures such as the gas–liquid interfacial area, which is ofgreat importance in the assessment of chemical processesin gas–liquid contactors. Figure 13 shows the extractedinterfacial area in an extra three-dimensional volume plot ofthe column.

Figure 12. Example of a dynamic image reconstruction. Inside thealuminium cup, three pearls are agitated by a stirrer rotating at 20rotations per second. The movement of the spheres is recorded withthe ultra fast electron beam tomograph at 1000 Hz frame rate. Fourrepresentative slice images out of a full sequence of 500 are shown.

9

Meas. Sci. Technol. 19 (2008) 094002 F Fischer et al

Figure 13. Axial cut and three-dimensional phase boundary plot ofthe gas phase in a bubble column. The sequence has beensynthesized from 500 slice images. Note that the vertical coordinatehas the dimension of time.

9. Conclusions

An ultra fast electron beam x-ray CT scanner for multiphaseflow measurement applications has been designed, builtand tested. The essential parameters are summarized intable 2. The scanner comprises an electron beam generatorwith 150 kV/10 kW high-voltage supply, beam focussing,centring and deflection system, a semicircular x-ray productiontarget made of tungsten alloy, and a 240-element CZT x-raydetector with a fast data read-out. In scanning mode, thebeam is circularly guided across the target to produce a rapidlymoving x-ray spot. The detector has a sampling rate of up to1 MSample s−1. After proper dark and reference calibration,the data are processed by standard computed tomographyalgorithms based on the filtered backprojection technique toproduce a time series of cross-sectional images. The maximumframe rate of the scanner is so far at about 7 kHz, limitedby the capability of the deflection coil amplifiers to adjustthe required elliptical beam deflection pattern. The spatialresolution in the images has been determined as 0.51 lp mm−1

at 10% MTF. Further, the noise floor in the images isabout 3% with respect to the mean attenuation coefficient ofwater. Scans on a dynamic particle phantom and a water–gastwo-phase flow have proven the functional principle of thescanner.

With scanning frequencies of up to 7 kHz, most practicalmultiphase flow problems in small geometries can be tackled.This includes the gas–liquid pipe flow, bubble columns,

Table 2. Main properties of the electron beam CT scanner.

Parameter Value

Electron beam gunPerveance (beam widening) 1.1 × 10−9

Max. acceleration voltage 150 kVMax. beam current 65 mAMax. beam power 10 kWWehnelt voltage 3 kVAuxiliary supply for electron 1 kV

bombardmentFilament heating current 8 AMax. deflection frequency 10 kHzMax. deflection angle ±17.5◦

Beam generating system RogowskiDimensions 650 mm length, 273 mm dia.Gas pressure 10−6 mbar

Detector and electronicsNumber of detector elements 240Size of detector elements 1.5 mm × 1.5 mm × 1.5 mmDetector material Cadmium zinc telluride (CZT)Maximum sampling rate 1 MSample s−1

RAM size 4 GB

Operating parametersMax. frame rate 7000 fpsFocal spot size 1.1 mmSpatial image resolution 0.51 lp mm−1 at 10% MTF

fluidized beds, flows in chemical vessels and many more.Phase fractions will be dynamically resolved up to flowvelocities of 5 m s−1 within the spatial and contrast resolutionlimits given above. Especially for scientific applications, suchas experimental investigation of multiphase flow with the aimof flow mechanical model development for CFD codes, thisscanner is a valuable new flow visualization tool. On the otherhand, the scanner may be used for accurate measurement ofphase fraction, e.g. for calibration of new multiphase flowmeters, since the dynamic high-speed imaging is not proneto typical dynamic bias errors known from time-averagingradiation measurement techniques [11].

References

[1] Boyd D P and Lipton M J 1983 Cardiac computed tomographyProc. IEEE 71 298–307

[2] Gould R G 1992 Principles of ultra fast computed tomographyUltrafast Computed Tomography in Cardiac Imaging:Principles and Practice ed W Stanford and J A Rumberger(Mount Kisco, NY: Futura)

[3] Johansen G A, Frøystein T, Hjertaker B T and Olsen Ø 1996 Adual sensor flow imaging tomographic system Meas. Sci.Technol. 7 297–307

[4] Morton E J, Luggar R D, Key M J, Kundu A, Tavora L M Nand Gilboy W B 1999 Development of a high speed -raytomography system for multiphase flow imaging IEEETrans. Nucl. Sci. 46 380–4

[5] Misawa M, Tiseanu I, Prasser H-M, Ichikawa N and Akai M2003 Ultra-fast x-ray tomography for multi-phase flowinterface dynamic studies Kerntechnik 68 85–90

[6] Hori K and Akai M 1993 Measurement of variation in voidfraction distribution by the fast x-ray CT scanner Graph.Simul. Vis. Multiphase Flow 21 96–114

10

Meas. Sci. Technol. 19 (2008) 094002 F Fischer et al

[7] Hampel U, Speck M, Koch D, Menz H-J, Mayer H-G, Fietz J,Hoppe D, Schleicher E and Prasser H-M 2005 Experimentalultra fast x-ray computed tomography with a linearlyscanned electron beam source Flow Meas. Instrum.16 65–72

[8] Bieberle M, Fischer F, Schleicher E, Koch D, A-ktay K S D C,Menz H-J, Mayer H-G and Hampel U 2007 Ultra fastlimited-angle type x-ray tomography Appl. Phys. Lett.91 123516

[9] Bieberle M and Hampel U 2007 Evaluation of alimited angle scanned electron beam x-rayCT approach for two-phase pipe flows Meas. Sci.Technol. 17 2057–65

[10] Kak A and Slaney M 1988 Principles of ComputerizedTomographic Imaging (New York: IEEE)

[11] Harms A A and Laratta F A R 1973 Dynamic bias in radiationinterrogation of two-phase flow Int. J. Heat Mass Transfer16 1459–65

11