An Open-Frame Loading Stage for High-Resolution X-Ray CT

Abstract

The utilisation of high-resolution in situ computed tomography (CT) in the (sub-)μm range is typically only viable in synchrotron facilities, as the deployment of a conventional loading stage in laboratory CTs with a cone beam source does not facilitate a corresponding geometric magnification. This publication presents a CT system with a novel in situ concept that allows spatial resolutions down to 0.5 μm, enabling the analysis of weakly absorbing materials capable of applying loads of up to 5 kN in both the compression and tension directions to the sample during the measurement. The necessity for a highly precise mechanical design to ensure successful measurements at magnifications approaching the theoretical limit makes the system’s development particularly demanding. The components employed are presented, along with the requisite considerations and methodologies. It can be demonstrated that the intended specifications with regard to precision and quality are met. The experimental results of a fibre-reinforced polymer demonstrate the system’s ability to detect matrix damage features below a single fibre diameter, thereby highlighting its potential for applications in materials science where traditional laboratory CTs are insufficient and synchrotron access is limited.

1. Introduction

The ability to generate high-resolution volumetric images of material samples is of great importance for a wide range of research questions in the natural and material sciences. X-ray-based computed tomography (CT) images of samples that are loaded and deformed during the measurement, i.e., where the measurement takes place in situ, are of particular importance. This allows the observation of deformation processes and the progression of potential damage before the final failure of the sample. This enables the specific conclusions to be drawn about the material behaviour that would not be possible if the material sample were examined post mortem [1,2].

Fibre-reinforced polymers (FRPs) comprise carbon, glass, or basalt fibres within an epoxy matrix and exhibit exceptional strength and stiffness. However, they are susceptible to damage mechanisms, including matrix cracks, fibre breakage, delamination, and voids. These microscopic defects, though initially difficult to detect, have a significant impact on the macroscopic material properties. The detection of such defects, particularly their initiation and growth under load, remains a key challenge in materials such as FRP with low X-ray absorption polymer matrices, as well as in biomaterials, natural fibres, synthetic polymers, and sustainable biopolymers [3]. This highlights the need for innovative laboratory CT techniques to overcome the experimental limitations and reduce reliance on limited-access synchrotron sources. Although synchrotron CT provides superior image quality and acquisition speed, laboratory CT systems are more practical for routine investigations, offering accessibility, cost efficiency, and the flexibility to accommodate larger samples for multiscale studies [4].

If a laboratory CT system, in combination with an appropriate in situ experimental setup, is capable of achieving sufficiently high spatial resolution to detect fine defects in the order of 1 μm within weakly absorbing matrices, it offers substantial advantages. By integrating in situ capabilities, laboratory CT systems can support the development of reliable composite structures, such as those required for large aircraft components. This ensures precise defect characterisation for structural integrity and optimal performance [5].

The aim of this work is to develop a laboratory CT system that fulfils the following requirements: the CT system should be capable of achieving a resolution of less than 1 μm and should be able to analyse samples of various materials, particularly those with low levels of absorption. Furthermore, the system should allow for the application of loads up to 5 kN in both the tensile and compressive directions while the sample is being measured, and it should be possible to continuously record the applied force. These premises necessitate specific requirements for the equipment used and the mechanical design, which are discussed in the following sections.

During CT measurement, the sample rotates around its own axis by up to 360° to generate X-ray images from different perspectives. For in situ mechanical loading, i.e., when a force is applied to the sample along the axis of rotation, a structure must exist which channels the reaction force around the sample. This structure is referred to as the load frame. One is typically not interested in imaging the load frame; instead, only the sample should be visible in the X-ray image. To achieve this, two different concepts can be employed, which are briefly introduced in the following (see also Figure 1).

Closed loading cage: A tube of a slightly absorbent material is placed around the sample. This tube redirects the reaction force of the applied force around the sample. As the tube remains permanently in the X-ray beam, it is important that its X-ray absorption is low. It is also important that its optical properties are homogeneous and invariant with respect to rotation such that the absorption occurs evenly over the entire cross-sectional area of the X-ray beam and remains consistent for each projection. For each projection angle, the entire structure, comprising the sample, the tube, and the load application mechanism, is rotated.

With the closed cage, a force can be applied to a sample with a simple design. However, the additional absorption adversely affects the measurement result. This is a particular drawback for samples which are themselves of low absorption and low contrast. Additionally, since the tube has a certain diameter around the sample, the minimum distance between the X-ray source and the sample is increased. This implies that the optical magnification in a cone beam X-ray setup is reduced, compared to the situation without the tube.

Open loading frame:The tube carrying the reaction forces can be dispensed by using a load frame outside the X-ray field of view. To obtain artefact-free measurements, a full rotation of the sample around its axis of rotation is required for conventional reconstruction algorithms such as FDK (Feldkamp–Davis–Kress filtered back projection method). Consequently, the load frame must not enter the X-ray beam at any time as stated in [6]. The load frame does not rotate; only the sample in its fixture is rotated.

In an open load frame, the contrast range of the X-ray system can be fully utilised, as only the sample has an absorbing effect on the X-ray beam. If the load frame is dimensioned accordingly, the sample can be moved right up to the window of the X-ray source to achieve maximum geometric magnification (by using cone-beam X-ray). However, the mechanical complexity of the open-frame setup is far more challenging than for the closed frame. As the closed frame simply induces gravitational forces on the rotation stage only, very precise air bearings can be employed, yielding high stability of the optical axis. This is a prerequisite for successful reconstruction, as the reconstruction algorithms typically assume a stable system, with the only variation from image to image being an increment of the projection angle. For the open frame, in contrast, the axial load on the sample has to be transmitted through the bearings. Here, for the intended range of axial loads up to 5 kN, this implies that air bearing are too weak, and that instead, ball or roller bearings must be employed. As will be discussed further below, this is a challenging aspect for the targeted resolution below 1 μm.

An important aspect of this work is that we deliberately focus on laboratory cone-beam CT. A number of CT results with resolutions of better than 1 μm exist in the literature; however, these are obtained in a parallel beam setup using synchrotron radiation [7,8,9]. While the properties of this type of radiation provide a brilliant measurement result, the availability of synchrotron sources is limited. Conventional X-ray tubes with conically emitted radiation are comparatively cheap and available at many research institutes [10]. The intention of this work is to demonstrate that sub-1 μm resolution CT is possible for in situ mechanical loading with cone-beam CT.

Existing Concepts for In Situ CT

Setups for the in situ measurement of different material samples in laboratory CT are described by a number of authors employing a closed load cage. Ref. [11] uses a simple load frame with a polycarbonate tube around the sample for the investigation of Al-Li-2090 samples. The tube has an outer diameter of 50.4 mm and a wall thickness of 2.8 mm, which leads to an attenuation of the X-ray beam by 35.6% at the characteristic line of the silver target (Ag $K_{\alpha }$ = 22.5 keV). In the CT measurement, a voxel size of 6.5 μm could be achieved with this setup. Germaneau et al. [12] investigate the deformation of copper-doped polymer samples in CT and use a load frame with a PMMA tube in the X-ray neck to carry the load. The strongly absorbing copper particles with a diameter of about 150 μm are visualised with a voxel size of 6 μm.

The deformation of a vertebral cover plate due to the load caused by an implant was analysed by Hulme et al. [13] using micro CT. For this purpose, an in situ loading device with a POM tube was used. The bone samples were equipped with several glass spheres with a diameter of 1 mm to set reference points for the deformation of the vertebral body. A voxel size of 82 μm was achieved.

Lachambre et al. [14] used a loading device with a load-bearing aluminium shell for micro CT and subsequent full-field digital volume correlation (DVC) for analysis of the sample deformation. With an outer diameter of 16 mm, the sample axis could be moved correspondingly close to the X-ray source such that a voxel size of 3.5 μm was achieved. The notched sample was made of spheroidal graphite cast iron. The contrast of the X-ray images was not sufficient to observe the progression of the crack; however, the residuals of the DVC algorithm were used for this purpose.

In conventional flat-panel detectors, the X-ray photons are registered immediately after hitting the detector’s scintillator. A noteworthy approach in recent research is to use optical methods to further magnify the photon beam after it has passed through the scintillator, which is possible because the beam is now in the visible light spectrum. This helps to overcome, to some extent, the limitation of the closed load frame as shown by Wan et al. [15] for a carbon fibre–silicon carbide ceramic matrix composite at a voxel size of 5 μm and Wang et al. [16] for a study of carbon FRP at a voxel size of 2.2 μm.

What all of these in situ experiments have in common is that a closed load cage was used. The effects of the polymer or aluminium tubes was usually not quantified in terms of influence on the X-ray image quality due to changes in absorption. However, it can be assumed that the tube contribution to the total absorption was negligible, as samples with relatively high absorption were imaged, which required the use of hard X-rays.

For in situ measurement of samples made of weakly absorbing materials, such as polymers and other organic compounds, however, tubes made of similarly or even more strongly absorbing materials are, a priori, not suitable. Besides the already discussed restriction of a reduced geometrical magnification, closed-frame loading cages do not appear suitable for high-resolution CT of samples with low X-ray absorption.

We were not able to find reports in the peer-reviewed literature for open-frame loading devices. A commercially available 20 kN open-frame in situ loading rig for CT imaging exists [17], which allows in situ measurement under compression, tension, and torsional loading. Due to the dimensions of the sample clamps, the minimum source–object distance is limited to around 45 mm. With additional efforts, the resolution could be increased to a voxel size of 14.6 μm in [18].

In the following, the design and construction of a CT system is presented which overcomes the described limitations of increased absorption and reduced magnification.

2. Materials and Methods

2.1. Requirements

The requirements for our CT with in situ testing capability are as follows: The laboratory CT system should be able to achieve a resolution of <1 μm. It should be able to examine samples of different materials, including those that are poorly absorptive. It should be able to apply loads of up to 5 kN in both tensile and compressive directions to the sample during the measurement. It should be possible to record the applied force continuously. From this premise, specific requirements for the equipment used and the mechanical design are derived and described below.

2.1.1. X-Ray Setup

The microfocus X-ray tube L10711-03 from Hamamatsu Photonics, Hamamatsu, Japan, is employed as the X-ray source. The tube voltage can be varied in the range of 20 keV to 160 keV. The tube is specified for a maximum resolution of 0.25 μm using a beryllium target of 0.5 mm thickness and an opening angle of the emitted cone beam of 140°.

Two different detector systems are employed. We use a CMOS flat panel detector for large-format imaging and the majority of applications, while a photon-counting detector is optimal for specific energy-resolved measurements. The flat panel detector is a Rad-icon 2329 from Teledyne Technologies, Thousand Oaks, CA, USA, comprising a silicon scintillator. The detector has a resolution of 4608 × 5890 pixels with a pixel pitch of 49.5 μm, a dynamic range of 3000:1 and a digitisation depth of 14 bits. For photon counting, we employ a Timepix series direct measuring semiconductor detector. The Wide-Pix detector used consists of 15 Timepix sensor elements and delivers a resolution of 3975 × 265 pixels at 55 μm pixel pitch. In TimePix detectors, incident X-ray photons can be evaluated for their energy or wavelength, which makes energy-resolved or spectral X-ray imaging possible [19].

2.1.2. Load Frame Design

The load frame is a classic design, where a stationary base and a movable crosshead are joined by two vertical linear guides. For an overview of the design concept refer to Figure 2 and Figure 3; a detailed view of the kinematics is provided in Figure 4. The base and crosshead are made from high-strength EN AW-7075 aluminium and house the drive components to effect vertical displacement and the rotary axis for CT measurement. The vertical displacement loads the testing sample and is realised using two symmetrically arranged ball screws (KGT-R-1605-RH-T5, Smalltec, Winnweiler, Germany), driven by stepper motors (drive torque 2 Nm, Leadshine ES-M32320, Shenzhen, China) with worm gear reduction (transmission ratio i = 20, worm gears H/I Mädler, Stuttgart, Germany). This allows to transfer a maximum force of 63 kN to the sample with a resolution of 0.83 μm per step.

The vertical guides are two precision ground rods of 40 mm diameter on which the crosshead slides using ball cage bearings (Fibro, Germany). The bearings are preloaded, to ensure zero radial play (pairing classification and preload class 2) and thus only axial movement.

The rotary axis which provides different projection angles is realised using super precision roller bearings with a sliding inner ring and reduced internal clearance (NN3005C1NAP4, NTN, Osaka, Japan). These bearings with tolerance class 4 according to ISO 492 [20] constrain the concentricity error, i.e., the radial run-out, of the inner ring relative to the outer ring to at most 3 μm. The axial run-out is constrained using a standard fixed bearing block (FK15, THK, Tokyo, Japan) with a tandem arrangement of precision angular contact ball bearings of the type B7002-E-T-P4S-UL. The running tolerances of these bearings are in accordance with tolerance class 2, which specifies a radial and axial run-out of the inner ring to 1.5 μm.

One FK15 bearing block is mounted to the load cell, which in turn is fixed to the base. The load cell is a ring-shaped force sensor with a nominal force of 5 kN (K3R, ME-Messsysteme, Hennigsdorf, Germany). In this way, the axial load can be monitored while the axis is allowed to rotate. The second FK15 bearing block is mounted to the movable crosshead. Each of the FK15 bearing blocks accepts a rotary axis, which is a precision ground rod with an integrated collet holder (HFER25M-Z25-L1=150, Fahrion, Kaisersbach, Germany). Commercially available precision shafts with a diameter of between 3 mm and 12 mm are employed as specimen holders. The specimen is typically glued to these shafts using adhesives. The shafts are clamped inside the collet holders with precision collets.

We note that air bearings, which are otherwise considered the gold standard for a stable rotary axis in microtomography, are no option for us. Air bearings are not suitable for the high static axial loads required here.

The rotary axis is driven by two servomotors (EMJ-04AFD22 motor with PRONET-04AMG servo drive, Estun, Nanjing, China). Their 20-bit encoder delivers a theoretical angular step resolution of $3.4\times {10}^{-4}$°, while the maximum achievable torque is 3.8 Nm. The servomotors share the same instruction signal line, ensuring synchronous operation to prevent the sample from twisting.

2.1.3. Mechanical Setup

Both the load frame and the detector are mounted on linear guides and can be moved along the optical axis using ball screws controlled by stepper motors. This allows to dial in different distances and thus different geometrical magnification factors.

Due to the high aspect ratio of the Medipix detector—it is essentially a line detector— rasterisation is necessary for two-dimensional images. For this reason, the detector is mounted to a motorised table, which allows the detector to be moved in the plane normal to the optical axis. The detector is connected to an adapter plate and can be easily interchanged with the flat panel detector. A reproducible connection is ensured by a positive locking mechanism of dowel pins.

2.1.4. Electromechanical Control

The entire control system is housed in a control box, with the stepper drives and servo drives connected to corresponding 24 V or 48 V power supply units. All mechanical degrees of freedom are controlled by using the Grbl motion controller [21], which accepts G-code instructions over a serial interface. A custom computer program written in Python facilitates the positioning tasks of detector and load frame before the measurement, as well as the movement of the rotary axis to achieve different CT projection angles. This computer program also controls image acquisition with the detector, starting exposure sequences and reading out the detector to the hard disk.

2.2. Cooling System

A necessary requirement for the successful reconstruction of projections to topographies is that both rotary and optical axes remain stable during the acquisition time, which can be on the order of several hours. It is therefore mandatory that heat generated by its electrical components is removed from the system to avoid thermal expansion and the associated changes in geometry. At the same time, the temperature within the X-ray tube must be kept constant to achieve a constant photon flux, which is necessary for constant exposure statistics of the specimen. While the latter requirement is necessary for CT at all scale, the former becomes critical for CT at the sub-μm scale because dimensional errors due to thermal expansion are measured relative to a very small length scale. This makes it necessary for us to build a custom cooling system, which is detailed below, starting with a numerical analysis of the thermo-mechanical behaviour.

2.2.1. Structural Effects Due to Thermal Expansion

The X-ray tube, detector, and load frame are components consisting of multiple materials with different thermal expansion coefficients. We require a prediction of how these components move relative to each other. This prediction is necessary to infer the maximum allowable temperature differences between the components, to limit geometry changes due to thermal expansion.

To this end, a thermal simulation was performed using FEM. This was possible because all parts of the system are modelled within the CAD software Autodesk Inventor, which contains the FEM solver Nastran. For this analysis, it was assumed that the optical table remains at a constant temperature, and the CT-system mounted to it is subjected to a temperature change.

We found that the dominant source of geometrical errors is due to thermal stresses arising from the aluminium structures mounted directly to the table: the table is constructed from a stiffened steel honeycomb sandwich structure comprising six millimetre-thick top layers and a core height of 200 mm. An 80 mm-wide aluminium profile is mounted on the table, with brackets at 100 mm intervals used to anchor it. The linear rails are fixed to the aluminium block. An increase in temperature would result in the aluminium profile expanding, which would cause the table top to bend and thus tilt the X-ray beam. The second-most dominant error term is the movement of the sample relative to the X-ray source, which results in changes in the magnification.

Using the FEM model, we determined the maximum permissible temperature change according to the following constraints: (i) The optical axis must not move more than one pixel within the detector plane, i.e., 10 μm. (ii) The source–object distance must change less than 1 μm. The simulations predict that, if the temperature of the structure increases by 2 K during the measurement period, the bending of the structure at a source–detector distance $SDD=400$ mm and source–object distance of $SOD=8$ mm would result in a vertical shift in the optical axis on the detector of 8.9 μm. This temperature change would also cause a horizontal expansion of SDD and SOD by $\Delta SDD$ = 8.9 μm and $\Delta SOD$ = 0.48 μm, which would lead to a change in the magnification ratio.

The geometric magnification m results from

$$m={\displaystyle \frac{SDD}{SOD}}$$

(1)

and thus by changing one of these variables to

$${\displaystyle \frac{\partial m}{\partial SDD}}={\displaystyle \frac{1}{SOD}}$$

(2)

and

$${\displaystyle \frac{\partial m}{\partial SOD}}=-{\displaystyle \frac{SDD}{SO{D}^2}},$$

(3)

this shows the dominant quadratic sensitivity of SOD. The magnification changes $\Delta m_{SDD}={\displaystyle \frac{\partial m}{\partial SDD}}·\Delta SDD=0.0012$, and $\Delta m_{SOD}={\displaystyle \frac{\partial m}{\partial SOD}}·\Delta SOD=-0.003$ leads to a total magnification difference of $\Delta m_{\mathrm{total}}=-0.0018$. The projected change in position P of a point at a distance of $d=1$ mm from the centre on the object side is therefore given by

$$\Delta P=\Delta m_{\mathrm{total}}·d=-0.0018·1\mathrm{mm}=-1.8 \mu \mathrm{m},$$

(4)

which may be considered insignificant compared to the pixel pitch of the detector, 49.5 μm.

2.2.2. Thermal Management of the X-Ray Tube

As X-ray sources convert less than 1% of the supplied energy into X-rays, a considerable amount of heat must be dissipated to the outside of the measuring chamber in order to ensure a stable photon flow [22]. A constant temperature of the X-ray anode is relevant for longer measurements, as the emitted photon current changes with a change in its temperature and also favours a migration of the focal spot [23]. A water cooling system was installed for this purpose, which is shown schematically in Figure 5. Driven by a circulation pump, the coolant flows through the X-ray tube (and if a detector with cooling connections is used, also through that) and out of the measuring chamber. The flow is then split, and part of it passes through a plate heat exchanger (30 plates, max. 66 kW, Wiltec, Eschweiler, Germany), which is connected to an external recirculating coolant flow of the laboratory ($T\approxeq 17°\mathrm{C}$). The constant temperature of the coolant is realised by a PID controller and a 3-way mixer, which regulates the mixing ratio of both flows. To avoid condensation inside the X-ray tube and the measuring chamber, the coolant temperature should be within the ambient temperature range. We use ($T_d=22 °\mathrm{C}$) as the target temperature. At a volume flow of ($\dot{v}=18$ L/min), the setup reliably regulates a temperature jump in the single-digit range to the target temperature within 3 s. However, with the smooth heat input from the X-ray tube and the detector, no temperature jumps are to be expected. The tube emits a relatively constant 50 W of heat.

2.2.3. Thermal Stability

Initial investigations with an uncontrolled cooling circuit and a closed measuring chamber without heat extraction showed that the measuring environment heats up to an unacceptably high level during longer measuring campaigns. At temperature increases of $\Delta T=5\mathrm{K}$ over a measurement period of 8 h, a significant drop in the photon current was also observed (up to 25%).

In contrast, with active cooling, the temperature and photon counts are stable with respect to time, once a period of equilibration has passed; see Figure 6. These data are from a recording of 6 h in duration. Sensors are attached at the locations T0–T5, c.f. Figure 5, in the measuring chamber. At the beginning of measurement, the X-ray tube is at room temperature. The tube is switched on (tube voltage 50 kVp, tube current 140 uA, anode current 10 mA) and starts to warm up. Over the first 100 min, a sharp rise in the temperatures of the X-ray tube and detector can be observed. After this equilibration period, all temperatures remain within approximately $\Delta T=\pm 0.25\mathrm{K}$. In particular, the tube temperature remains stable within $\Delta T=\pm 0.1\mathrm{K}$.

In addition to the temperatures, Figure 6 shows the photon flux intensity as a function of time. From 100 min onwards, the flux is stable with a slight increase of less than 3% over 4 h. It is also important to acknowledge that the stability of the photon flux is influenced by additional factors, such as the condition of the X-ray target.

The present cooling concept is thus sufficiently functional for long measurement campaigns with the CT system. The temperatures are constant with fluctuations within the limits shown, which means that no significant measurement inaccuracy or artefacts due to thermal expansion of the structure are to be expected as outlined in Section 2.2.1. Moreover, the photon flux is sufficiently constant (with a deviation below 5%) for extended CT measurements; see [24].

2.3. Concentricity and Stability of the Rotation Axis

To determine the run-out, a 6 mm test pin of tolerance class 0 according to DIN 2269 (roundness: 0.5 μm) is inserted into the upper and lower sample holders, i.e., the collet chucks. Both the concentricity tolerance and the axial run-out are determined in several rotations at a distance of around 10 mm from the collet using a mechanical test indicator (MarTest 800 SGE, Mahr, Göttingen, Germany, 1 μm scale division). The data in

Table 1. Radial and axial run-outs of the rotary axis.

	Run-out/μm	Axial Error/μm
Top	1	3
Bottom	<3	3

show the run-out to be ≤3 μm.

For an independent measurement, the rotation axis is quantified from image projections using an analytical approach [25]. A high-precision steel sphere with a dimensional and shape tolerance of 0.08 μm (diameter 0.8 mm, tolerance class G3) is mounted to be rotationally eccentric on a sample holder. Projections at 360 discrete angular positions within one full rotation turn are collected with the X-ray system. The trajectory of the sphere’s projections from the sphere’s centres of gravity results in an ellipse whose minor axis becomes larger the further away it is from the X-ray source–detector optical axis. By analysing the trajectories at different vertical distances of the steel sphere, the tilting of the system around the optical axis can be precisely determined. The distances of the optical system, source–detector distance, and source–object distance are determined and calibrated by measuring points at different distances and by utilising the sphere diameter.

2.4. Positioning Repeatability of SOD and SDD

The positioning accuracy and, in particular, the repeatability of the SOD and SDD positions, are determined using a mechanical feeler lever measuring device, with a result of less than 10 μm. It should be noted that due to the backlash of the ball screws, accurate positioning is only guaranteed if the approach is always from the same direction. Automatic backlash correction is implemented in our control software environment by ensuring that the last section of a positioning move always occurs along the same direction.

2.5. Accuracy of Force Measurement

To validate that the force measured by the load frame’s force sensor corresponds to the force applied to the specimen, an axial force sensor is clamped between the collets. This sensor was previously calibrated in a universal testing machine of the 0.5 accuracy class. Five series of measurements were carried out with different load speeds and cycles up to 1200 N in tension and compression. The force acting on the sample differed from the reference by less than 5%. This difference can be explained by the bearing design: the roller bearings fixing the radial position of the rotation axis permit axial movement, but they require a few turns of rotation to completely eliminate friction. This is the case during a CT measurement but not during the static situation considered here.

2.6. Two-Dimensional Resolution of the System

The combination of the fine focus tube and the detector allows for a theoretical limiting resolution of 0.25 μm (focal spot size in high resolution mode). During a CT measurement, the sample is rotated around its axis in the sample holder, and a safety distance from the X-ray window must be maintained to protect the sensitive beryllium target from damage in the event of sample rupture. Therefore, a minimum distance of SOD = 3 mm is considered to be a practical minimum. A sample cross-section of 2 mm allows for the realistic theoretical spatial resolution of 0.4 μm to be achieved (geometric data: detector width 5890 px, SDD = 437 mm). To prove the actual ability of the system to achieve a sub-micrometer spatial resolution, a 2D test target (XRN-SCT-0054, XRnanotech, Villigen, Switzerland) is imaged, which contains line pair pattern with distances between 0.4 μm and 50 μm. In Figure 7, the intensity distribution along the vertical red line is plotted across the 0.7 μm lines. For this purpose, 10 pixels are averaged in the direction of the stripes. It can be shown that the 5 line pairs of the 0.7 μm structure can be clearly distinguished from the background noise. As a benchmark, the 5 minima of the 400 nm line group can still be significantly differentiated in this setting (green line). The following measurement parameters are used: X-ray voltage 80 kV, X-ray current 148 μmA, 120 measurements averaged at an integration time of 500 ms, SDD 840 mm, and SOD 4 mm. A sub-μm spatial resolution can therefore already be realised in the 2D image [23].

3. Results and Discussion

Validation of the setup was carried out with a basalt fibre-reinforced polymer that had already been characterised and measured in other systems [26]. It is a quasi-isotropic multilayer composite that consists of eight layers, ${[0/+45/-45/90]}_s$, with the outer two 0° layers running parallel to the load direction and the inner two layers orthogonal to it. The sample with a cross-sectional area of 1.6 mm × 2 mm was glued into bores of precision shafts with a diameter of 6 mm and clamped in the sample holder of the load frame. The use of thermal activated epoxy adhesive LOCTITE EA 9514 in 20 mm deep bores provided sufficient holding force for tensile testing. This resulted in an adhesive surface area of 144 mm², with the maximum shear strength of the adhesive being 44 MPa. Consequently, a force transmission of over 6 kN was thus possible. The CT settings and parameters used here are listed in

Table 2. Parameters of the validation measurement of a basalt fibre-reinforced polymer.

X-ray configuration	Tube voltage	50 kVp
	Tube current	140 μA
	Target current	10 μA
Acquisition settings	Integration time	5 × 1 s
	Projection angles	1000
Geometric configuration	SOD (source–object distance)	5.58 mm
	SDD (Source–detector distance)	398 mm
	Magnification	71.2
	Detector pixel size	49.5 μm
	Effective pixel size	695 nm

. The geometric magnification led to a geometrically native resolution of 0.7 μm, which corresponded to the realistically attainable resolution of the system as shown in Section 2.6. The first measurement was carried out in a load-free state. Subsequently, tensile loads of different magnitude were applied, and CT imaging was performed at each loading stage. The reconstruction of the image data was carried out using an OSEM-type iterative algorithm from a commercial reconstruction toolbox [27]. Figure 8 shows three sectional images of the reconstructed volume. The highly absorbing fibres are clearly and continuously displayed, free of imaging defects and artefacts. Artefacts in the volume such as ring artefacts are hardly noticeable. Beam-hardening effects due to the heterogeneity in the absorption of the materials can be surmised in the corners of the top view. The weakly absorbing epoxy matrix is also resolved with sufficient contrast. It is remarkable that even defects such as air pockets (detail i) and damage to the laminate due to the increasing load can be inferred from the image (interfibre break or microcrack; detail ii). The interfibre fracture occurs as the load on the sample increases as can be seen in Figure 9.

The intensity distribution along a line through the reconstructed volume, shown in Figure 10, provides information about the resolution of the system under these test conditions. This so-called edge spread function shows the jumps from weak absorption (matrix) to maximum absorption (fibre). Over an average distance of ≊3.5 px, the contrast increase is ≊3000 intensity values. Given the pixel-to-length ratio of 0.7 μm, this implies that individual features smaller than 2 μm can be recognised in the image, underlining the capability of the setup to achieve good-quality imaging for difficult materials such as fibre-reinforced polymers.

In addition to interfibre breaks and air inclusions, the volumetric rendering in Figure 11 shows details such as fibre breaks at the 0° position, highlighting the particular potential of this in situ X-ray system. No fibre breaks are found in renderings below 75% of the expected universal tensile strength.

4. Conclusions

This work demonstrates the suitability of the open-frame loading frame type for high-resolution in situ microtomography. The implementation presented here achieves a native spatial resolution of 0.7 μm in projection images. Sample CT data for a fibre-reinforced material show that the device is capable of identifying minute damage and fracture details at a scale well below a single fibre diameter, i.e., individual features with a size of ≊2 μm can be recognised.

The open-frame construction allows for maximum geometrical magnification, as the X-ray source can be placed directly in front of the object. This is in contrast to the closed-cage type of in situ loading devices, which is placed on the rotation stage of an existing CT setup, such as the well-known stages from the manufacturer Deben, UK. However, the open-frame design comes at a price: as the rotation axis also has to transmit axial forces, the use of high-precision air or magnetic bearings is prohibitive, as these cannot carry the high axial loads of 5 kN required in our case. Instead, roller bearings have to be employed. Our design uses a bearing arrangement similar to high-precision machining spindles, with dedicated axial and radial constraints of the degrees of freedom. We show that this design works satisfactorily for the intended use of the in situ computer tomography of heterogeneous materials, such as the here presented results for a fibre-reinforced composite. A second advantage of the open-frame design is that the X-ray spectrum only needs to pass through the object, without passing through any additional material, as is the case with closed loading stages. This allows the investigation of materials with very low absorption characteristics, as the X-ray spectrum is not filtered.

The following represents an estimation of the financial outlay incurred within the European market. The total cost of the system is constituted by the sum of the individual parts, the precision mechanical work, and the design and development work. The cost of an X-ray tube ranges from EUR 100,000 to EUR 150,000, while a comparable flat-panel detector costs approximately EUR 50,000. Furthermore, the remaining components and workshop costs are insignificant, amounting to approximately EUR 10,000 each. While commercial CT systems (priced between EUR 0.8 million and EUR 2.2 million) frequently necessitate substantial maintenance expenditures, an experimental system such as the one under discussion here requires a comparison of the personnel costs associated with production and maintenance.