Challenges in non-destructive X-ray CT testing of riveted joints in the automotive industry

Abstract

Despite recent advancements, contemporary laboratory-based industrial computed tomography (CT) technology has not yet achieved the status of an efficient and economically viable in situ non-destructive method for examining riveted joints that are commonly used in the automotive industry. 2D X-ray or 3D CT systems can only reliably assess a rivet joint’s condition in rare, and favorable circumstances. Moreover, a comprehensive analysis to identify the obstacles preventing CT from efficiently capturing high-resolution, and contrast-rich scans of riveted joints in general is lacking. This work delves into the challenges and limitations of the technology that prevent operators from reaching the desired precision required for rigorous quality control. The study specifically addresses questions concerning in situ reachability and positioning of the sample structure, radiodensity and beam hardening, the relationship between spot size and power, and economic feasibility—all within the context of analyzing riveted joints in the automotive industry. Additionally, we provide an abstracted overview of the current industrial X-ray tube market, which informs our discussion on these specific aspects. Through a combination of experimental findings, analytical insights, and deductions, we shed light on the persistent difficulties in achieving resolutions comparable to macro-sectioning and microscopy, which are currently considered the gold standard in examining riveted joints. Our investigation further explores the potential of a photon-counting detector paired with a conventional, laboratory-based X-ray source. Furthermore, we investigate the use of a synchrotron beamline as an X-ray source in conjunction with a custom-designed detector optimized for such beam geometries. This detector employs scintillators coupled with optical magnification, functioning similarly to an X-ray microscope. Our findings and data collection conclusively reveal that commercially available CT systems still fail to meet the requirements for efficient testing, underscoring the urgent need for further research and development to enable widespread implementation.

Article Highlights

• Highlighting the limitations for economically assessing riveted joints using computed tomography.

• Comparing traditional CT scans with a synchrotron-based radiation source and a photon counting detector.

• Abstracted representation of commercial industrial X-ray tube market.

1 Introduction

Car chassis and bodies are typically composed of an array of components, including frame rails, cross members, and body panels that are securely interconnected with a combination of mechanical fasteners such as screws and rivets, as well as non-mechanical bonding techniques such as welds or adhesives. The number of each joining technique in cars may vary depending on the make, design, material, and construction of the vehicle.

The structural quality of mechanical joints, for example self-piercing rivets (SPRs), is predominantly evaluated via macro-sectioning (see, for example, [20, 22, 27, 29]). Macro-sectioning, while effective, is characterized by a somewhat intrusive nature that damages the subject being examined (Fig. 1a, b). Furthermore, it offers only limited two-dimensional insight (Fig. 1b) and is reliant on a series of laborious manual steps [28, 63]. To be more specific, and in proper chronological order, macro-sectioning involves the following stages: (1) extraction of the mechanical joint and portions of its structural context from a larger object, like a car frame, (2) meticulous dissection into two halves, (3) precise grinding and polishing, (4) chemical post-processing,¹ culminating in (5) microscopic evaluation of the joint’s cross-section. At this final stage, critical aspects such as undercut/interlock, head protrusion, and die-side material thickness are scrutinized. Throughout the entire undertaking, great care must be taken to avoid distorting the joint by way of excessive force application. It should also be noted that the accuracy and effectiveness of the procedure somewhat hinge upon the proficiency level of the person performing it.

This shows that analysing rivet joints in conventional car bodies requires extensive time and resources to test for strength and assess their structural quality after processing, as well as the unfortunate fact that strength and quality evaluations cannot be performed on the same sample, let alone comprehensively, due to the destructive nature of macro-sectioning.

Fig. 1

Depiction of destruction of both the initial object (the car frame) and the joint itself

X-ray computed tomography (CT) is a powerful imaging technique that is highly effective in capturing detailed structural information in a non-destructive fashion. In comparison to other methods, such as thermography, eddy-current testing, terahertz, or ultrasonic imaging, X-rays can provide orders of magnitude more detailed structural information, making it an essential tool in many fields such as materials science, non-destructive testing, and medical imaging. Conventional, laboratory-based, industrial CT systems employ X-ray tubes that offer acceleration voltages from 60 kV up to 600 kV [54, 57,58,59, 61, 62]. Conventional CT systems are predominantly equipped with an energy-integrating flat-panel-detector (FPD) or line-detector-array (LDA). Radiation sources that operate in the MeV (mega-electron volt) range, such as radionuclides ($\gamma$-radiation) or linear accelerators (LINACs), are not discussed here due to their potential to induce radioactive decay, and therefore require a significant waiting period to allow for the radioactivity to decay to safe levels.² Due to their inherent and heightened risks, radioactive methods are oftentimes not favored, particularly within the realm of producing industries where human contact must be avoided. In addition, the utilization of high-energy radiation sources requires a substantial increase in radiation protection measures, which inevitably results in additional costs. Thus, for the present purposes, X-ray radiation sources with lower energy levels are preferred.

Although the pursuit of non-destructive testing of steel riveted joints, or similar structures, is not a new endeavor, and the feasibility of using X-rays on such samples has long been investigated [38], a commercially viable and technologically feasible solution has yet to be developed. Furthermore, crucial facets have been overlooked or left unexplored, like the intricacies of voltage-dependent beam hardening artifacts and their potential impact on the detectability of fine structures [21]. While some observed and acknowledged the degrading influence of strong artifacts [19] they did not disclose essential X-ray parameters such as voltage or magnification, nor did they specify the system used. This makes a scientifically comprehensible classification of the results and the associated scope for optimization difficult to quantify. Furthermore, there are others who have demonstrated that sub-micrometer resolution with CT is indeed realistically achievable [13]. However, their findings can only be extrapolated to relatively small samples exhibiting low radiodensity—a measure of how well X-rays pass through an object. Thus, they cannot be directly applied to typical specimens found in the automotive industry, which are significantly larger, heavier, and denser than the samples investigated in the aforementioned study.

Although the technology has undoubtedly made considerable progress over the last decades in terms of resolution, accuracy, and speed, for many use cases its limitations and challenges remain obstacles to achieving the desired level of precision and automation necessary for rigorous quality control in an industrial setting. While there are some examples of implementations that yield data of sufficient quality for certain samples, defects, and use cases, a fundamental analysis has not been conducted to determine the obstacles preventing CT from efficiently capturing truly non-destructive, high-resolution, and contrast-rich scans of riveted joints. The methodology should strive to achieve a scan quality that is on par with the currently widely used destructive method that involves macro-sections and microscopy, all while ensuring its feasibility within an industrial environment. The aim of this work is to demonstrate that, despite recent advancements, contemporary laboratory-based industrial CT technology does not yet offer a universal, sufficiently reliable, efficient, and economically viable non-destructive method for examining riveted joints of the kind that are commonly used in the automotive industry. We provide a comprehensive discussion of relevant technical and physical aspects within the CT-imaging chain, examining their current technological capabilities, inherent benefits, and limitations. Although some limiting aspects of that imaging chain may seem to have technological solutions to overcome them, a combination into a unified system poses significant engineering difficulties, operational inefficiencies, and high maintenance requirements, rendering it impractical for industrialized operations.

2 Factors impacting quality in industrial X-ray CT scans

The precision and accuracy of X-ray imaging data are influenced by a myriad of factors. It is essential for the operator to exercise an awareness of the prevailing limitations of the CT scanning equipment, as well as the effects that each individual scan parameter has, both independently and cumulatively, upon data quality when assessing objects for analysis. Specifically and to ensure optimal results, the scanning strategy, the type of X-ray source, and the detector must be carefully evaluated, as their suitability is use-case dependent. The subsequent section aims to provide a comprehensive overview of relevant factors and technologies, clarifying their impact on image quality and delineating their inherent potentials and limitations. Furthermore, it serves as a basis for the ensuing discussions in Sect. 4.

2.1 X-ray generation

The choice of technology behind the X-ray generation has a great influence on both the acquisition time and the achievable scan quality. Today, the most advanced form of generating X-rays is based on deflecting or accelerating charged particles³ over long distances in a vacuum. Examples for such systems are synchrotron radiation sources at storage rings, linear accelerators (LINAC), and free electron lasers (FELs). In the field of X-ray generation, these systems offer unrivaled resolution and material contrast, primarily as a result of the ability to monochromatize the spectrum with photon energies up to several hundred kV. However, they are typically limited to efficiently imaging tiny samples, ranging from millimeters to a few centimeters, due to the comparatively small field-of-view (FOV). Additionally, such facilities are considerably larger (several hundred meters in circumference or length, respectively) compared to laboratory-based systems [32, 41, 42].

Standard, laboratory-based X-ray sources, on the other hand, exploit generation of X-rays when electrons interact with matter. Such systems are referred to as X-ray tubes. In contrast to synchrotron facilities, impact-based X-ray sources possess a polychromatic spectrum, encompassing both characteristic and bremsstrahlung radiation. When considering these systems in an industrial context, the maximum acceleration voltage, $U_{\text {acc}}$, as well as the related beam power, and focal spot diameter, are the most relevant tube-specific parameters. These three parameters provide crucial insight into whether the X-ray source is fit-for-purpose for the particular sample-specific use case. A deeper analysis of this subject follows in Sects. 3.2 and 3.3.

There are different standard tubes with characteristic focal spot sizes, peak acceleration voltages (kVp) and comparatively large FOVs. Corresponding to the size of their focal spot size, they are referred to as nano-, micro-, meso- and mini-focus tubes. There is always a trade-off between being capable of spatially resolving small features and penetrating heavy, dense, or large⁴ objects. This has to do with the power density $\left[ \text {power}\big /\text {area}\right]$ that is limited by the anode material’s heat dissipation, the intended magnification, and the resulting image blur. Various technological solutions have been developed to keep the spot size small at high energies and thus guarantee higher resolutions. For example, employing anode materials that have particularly good thermal properties or anodes that are actively cooled by liquids.

Rotating anodes are common for medical applications. They guarantee high currents and correspondingly good thermal conductivity. In medicine, high currents and the resulting short exposure times are preferred to minimize possible patient motion artifacts. However, much higher resolutions are usually required in industrial than in medical applications. Therefore, the majority of industrial X-ray sources are not rotating anodes. Recently, however, rotating anodes with a small, stable focal spot and different designs (450 kV, 225 kV) have been introduced [52, 53, 55]. The coming years will reveal whether this concept gains traction in the industry, considering its anticipated intricate and hence more expensive maintenance. Consequently, the pivotal question arises as to whether the advantages offered by these systems outweigh the financial disadvantages. The manufacturer asserts that economic feasibility is only attained for applications with a significantly high sample throughput, highlighting their commitment to delivering efficient solutions [64].

The Liquid Metal Jet (LMJ) anode [34] offers yet another way to reach high power densities. This concept uses a liquid, highly pressurized metal jet as a target that is made of an eutectic alloy (gallium, indium, tin) called Galinstan. This approach enables comparatively high power densities with small focal spots (1000 W at $30\,\upmu \hbox {m}$ focal spot diameter [56]). However, the current availability of such sources is limited to 160 kV [56], which may prove insufficient in meeting the energy requirements of numerous industrial applications. These include the casting of materials or the production of electrical components such as battery cells or electric generators that incorporate substantial quantities of copper or other high-Z elements.

2.2 Digital detector

The core element of an energy-integrating indirect X-ray detector is the scintillator crystal and the underlying photodiode array with a set pixel-pitch. Opting for a small pixel-pitch, and therefore higher resolution, necessitates the use of a thin scintillator to mitigate the effects of scattering within it. However, thinner scintillators lead to a lower quantum efficiency. This trade-off results in longer scan times and illustrates the inherent compromise between scan efficiency and achievable image detail. Direct X-ray detectors, or photon-counting detectors, circumvent the conundrum associated with scintillators by eliminating their use altogether. This not only allows for smaller pixel-pitches and consequently enhanced resolution but also enables superior material contrast owing to their distinctive capacity for energy discrimination.

A direct X-ray detector with both a large FOV and small pixel-pitches, however, remains computationally elusive.⁵ Such a combination would result in a tremendous volume of data per projection and, by extension, per scan. Contemporary limitations in data transfer rates and storage capacities render this aspiration impractical. Consequently, direct X-ray detectors are, at present, typically smaller in size compared to their conventional indirect counterparts.

The work of Sjölin et al. [9] captures the strides that photon-counting detectors have made in the realm of medical applications in the past decade. Their analysis underscores the pivotal juncture at which the market for clinical systems presently finds itself. Industrial applications, for example battery inspection in the automotive industry, emerge currently. A contribution of Dreier et al. [10] demonstrates how a combination of liquid metal jet and photon-counting detector can be used as a high speed, in-line inspection method for battery inspection. The work of Masuch et al. [30] demonstrate inspection techniques using X-rays for quality assurance of lithium-ion battery manufacturing in the laboratory or in large-scale production. The authors encountered challenges in visualizing the electrolyte sufficiently well, but express optimism about the potential solution offered by the latest developments in X-ray imaging. Specifically, they mentioned photon-counting detectors as a promising technology that could address this issue effectively. Currently, industrial applications of photon-counting detectors that closely align with the use case of this publication are either in the early stages of development or lack publicly accessible information. This scarcity underscores the urgent need for more comprehensive investigations in that field. With all that in mind, and depending on other imaging parameters like sensitivity, focus detector distance (FDD), acceleration voltage, tube current, and image acquisition time, a suitable detector solution for the case at hand must be considered.

2.3 Imaging parameters

Contrary to the relatively consistent nature of biological specimens examined in medical CT scans, the spectrum of industrial objects subject to analysis is a lot more diverse. These objects may differ significantly not only in size and geometry but also in material composition and quantity as is shown for example in [8].

As with any imaging system that has to be able to image a wide range of different objects, computed tomography systems offer a correspondingly large set of imaging parameters. They can be adapted to the individual use-case and to the nature and material composition of an object in order to improve the scan quality and remedy imperfections [24]. Improving the scan quality can be achieved by increasing resolution, increasing the contrast between different materials and/or densities as well as reducing CT-specific artifacts like beam hardening, photon starvation, scattering, or ring artifacts. All the above can be achieved by tuning the respective parameters [24] like magnification, acceleration voltage, current, physical filters, number of projections etc.

Due to the large number of parameters, finding an adequate combination can be quite time consuming [39] and may result in a trial-and-error approach. In addition to that, the amount of time for the scans themselves can easily reach the double-digit hour range. In an industrial context, particularly when production is impacted, longer investigation times indirectly result in increased financial losses. The time required for finding the set of parameters and the scan-time itself must therefore be kept to a minimum.

In order to find an adequate object-specific parameter combination, the operators of a CT system oftentimes have to rely entirely on their expert knowledge, longstanding years of experience, and established heuristics. Some even argue that optimal imaging parameters cannot be derived by analytical means [1], while others propose a knowledge-based system [40] to retrieve parameters from previous scans of similar objects.

2.4 Scanning strategy

The concept of computed tomography dictates that in order to volumetrically reconstruct an object as accurately as possible, the object must be transradiated from many different perspectives [11, 36, 43, 46]. During the image acquisition phase the detector and the source are typically diametrically arranged towards each other with the sample between them.

Circular trajectories are the most common and easy to implement. Source and detector rotate around the sample with a constant distance (FDD) to each other. The position of the sample (focus to object distance or FOD) does not necessarily need to be in the center of the aforementioned circle. For practical reasons it is common in industrial CT to have the object rotate on a manipulator table instead of the source and detector, like it is the case in medicine. Linear movement of source and detector (perpendicular to FOD and FDD) is used to increase the field of view horizontally and vertically. Relative movements of source, detector or sample in the direction of FOD and FDD is used to adjust magnification, and therefore resolution, as well as the amount of radiation (described by the inverse-square-law) on the detector. In many cases, a $180^{\circ }$ half orbit plus the cone-beam angle may suffice in terms of quality and detectability. However, in cases where noise and artifacts dominate, it is advisable to complete a full $360^{\circ }$ orbit.

Helical CT scans have been widely recognized as a standard practice for reducing cone-beam artifacts⁶. For instance, Samber et al. [7] demonstrated the reduction of cone-beam artifacts by scanning a cylindrical battery cell using a helical trajectory, which allowed for vertically capturing a wider angular range. This approach not only diminishes artifacts at the circular cone-beam image’s vertical extremities but also utilizes different rows of detector pixels for slice reconstruction, effectively mitigating detector-related issues such as ring artifacts. For rivets that are processed in car body parts, however, helical trajectories usually are limited by the material that surrounds the rivet.

High-resolution three-dimensional reconstructions of objects with large aspect ratios, such as thin, sheet-like workpieces including circuit boards, metal sheets joined by rivets or even prismatic battery cells, present a significant challenge for X-ray CT. The difficulty lies in the detection of minute flaws within these objects. A promising solution to this problem is the implementation of adaptive or variable zooming. As an example, Nikishkov et al. [33] have developed an innovative technique that leverages this concept. Their method employs a scanning trajectory, which adjusts the sample’s x-y-position based on the angular position, coupled with a novel reconstruction weighting strategy. Through this technique, they have demonstrated the ability to detect smaller defects in composite panels that have been impacted by low-velocity loads that conventional CT methods failed to identify. The work of Nikishkov et al. underscores the potential of variable zooming to revolutionize the inspection of large, flat objects without destroying them. This novel approach to X-ray CT, which involves continuously changing the FOD mid-scan and applying specialized weighting during reconstruction, is now a purchasable option in some of the latest commercially available, industrial CT-systems. However, in some cases additional hardware (motorized axes) as well as software (for the reconstruction) is required for such trajectories.

Free trajectories can be realized in two ways: (1) The source and detector remain stationary, while the sample is affixed to a suitable robotic arm [35] that grants the freedom to precisely position the sample between the source and detector. Alternatively, the sample can also be mounted on a hexapod [3], although this option imposes certain limitations on the range of possible sample alignments when compared to the robotic arm. (2a) The sample does not move. Source and detector are mounted on a c-arm like structure [31] that can be positioned freely around the sample. This has the advantage that source and detector are sufficiently well aligned in every position. However, the rigid connection between source and detector limit the accessibility for ROIs on structures like an automotive chassis or body in white (BIW). (2b) The sample does not move. Source and detector are each mounted on their own robotic arm [4, 49,50,51], see for example Fig. 2. Such a twin robotic system has the advantage that the physically not linked source and detector can move more freely around structures and perform more complete and optimized trajectories. However, due to the missing physical link between source and detector the correct position and orientation of their respective centers cannot be guaranteed during the scan. The magnitude and impact of this potential source of error presents a measurable problem for such systems [16]. In this case, positional as well as orientational calibration methods need to be applied: In [47], for example, the authors managed to correct for the central source-detector-offset by scanning a mathematically known calibration body before scanning the actual object. By exploiting the high repeatability of the robotic arms, the authors managed to correct every sample projection with a direct linear transformation based on the shift of the calibration body projection. The authors showed, that their method improved the scan quality extensively. They also showed that correctly assigning phantom markers might depend on the point of view in the projections. The work of [48] tries to remedy this with a machine learning based approach. The authors from another endeavor [26] proposed and quantified a method with a novel 3D-printed phantom that can be used for calibrating nearly any view point of a twin robotic CT system. A successful task-specific trajectory optimization on a robot-based industrial CT platform has been performed [2], demonstrating that such methods can be practically applied in a shop floor environment. The optimized trajectory resulted in a reduction of required projections by 55% or an improvement in image quality by 40% compared to conventional planar acquisition.

Fig. 2

Example of an industrial robotic CT system

3 Case study of CT inspection of riveted joints in the automotive industry

In the realm of industrial NDT analysis, selecting a CT system and its scan parameters hinges on four fundamental questions, each aimed at ensuring the setup is fit for the intended purpose. The prioritization of the subsequent questions may vary between different objects and industries:

• 1. Pre-scan preparation and positioning Can spatial accommodation, accessibility of the region of interest, and the secure stabilization of the sample be confirmed?

• 2. Acceleration voltage What is the optimal acceleration voltage $\left( U_{\text {acc}}\right)$ for the CT system to achieve sufficient penetration of the object?

• 3. Spot Size and Power: At the chosen acceleration voltage, what is the highest tube power $\left( P_{\text {max}} = U_{\text {acc}} \cdot I_{\text {max}}\right)$ that can be applied without compromising the desired resolution or damaging the system?

• 4. Acquisition time Given the selected $U_{\text {acc}}$ and spot size, can the CT system perform the scan within an acceptable timeframe?

It is imperative for practitioners to ascertain the necessary resolution before engaging with these setup-queries and selecting an appropriate CT system. Within the automotive industry, the precise dimensioning and qualification of rivet joints hinge on the clear visibility of critical features (for example the contact points between the rivet and both the upper and lower plates). Such features frequently possess dimensions within the single-digit micrometer range, underscoring the necessity for high-resolution imaging to guarantee the structural soundness and good quality of the rivet connections..

Upon setting the voltage $\left( U_{\text {acc}}\right)$, spot size, and acquisition time, it may become necessary to fine-tune additional parameters (refer to Sect. 2.3) to account for their intricate interdependencies. While questions 1, 2, and 3 must be fulfilled as essential criteria, question 4 addresses the scan’s efficiency and, consequently, its economic viability in the context of industrial non-destructive testing. The subsequent sections encompass a combination of experimental findings, analytical thoughts, and inferences. Each section corresponds to one of the above questions in a sequential order, ensuring a cohesive alignment between the questions and the nature of the content within the respective sections.

3.1 Constraining influence of sample geometry and its nature on pre-scan preparation and positioning

As discussed in Sect. 2, the typical setup in most industrial CT applications involves positioning the sample between a linearly movable X-ray tube and detector, ideally allowing for an rotation of $360^{\circ }$. To preserve the integrity of automotive rivet joints and minimize the influence of external forces, it is preferable to conduct in situ scans without detaching them from the car body. The preferred sample geometry for CT analysis is a slender cylinder composed of only moderately dense material. However, this poses a significant challenge when dealing with large body parts or fully assembled bodies, as they often deviate from the optimal cylindrical geometry and material composition. In the case of conventional CT systems advanced trajectories like variable zoom can prove beneficial (see Sect. 2.3). Careful adjustment of the FOD and FDD is necessary to ensure adequate magnification and photons reaching the detector. While both FDD and FOD can impact the resolution of the scan, it is important to note that only FDD directly affects the photon count at the detector. Therefore, improving FDD may require a corresponding adjustment that could potentially compromise the resolution. Regrettably, when dealing with rivet joints from the automotive industry, achieving acceptable quality with conventional laboratory-based CT necessitates separation and cutting of the joints. This process can potentially alter the joint’s structure and add an extra step to the analysis process chain.

A promising solution to address the positioning challenge is the utilization of robot-based computed tomography. Scientific and industrial examples of such systems have been discussed in Sect. 2.4. Such an advanced system offers the ability to access areas, inaccessible for conventional CT systems (see for example Fig. 2a, b), providing maximum trajectory freedom. Consequently, it can adaptively follow the most suitable trajectory depending on the sample being examined. However, due to the increased freedom of movement of the tube, detector, and robots, it is crucial to exercise caution to prevent collisions when employing trajectories that closely encircle the sample or body. The agility of robotic arms, as they weave in and out of the chassis, necessitates a well conceived selection of both the detector and the X-ray source. It is imperative to opt for models that are compact in design, yet meet the specific requirements of acceleration voltage and active detector area for the task at hand. One must exercise caution when considering the use of a LMJ tube in such applications. Its unsuitable dimensions present a significant limitation. Moreover, the LMJ tube’s high-pressure metal jet and inflexible tube system pose additional challenges. In the unfortunate event of the tubes being inadvertently bent and ruptured, the leakage of liquid metal (galinstan), could inflict damage on nearby structures. This is due to gallium’s corrosive properties when in contact with certain solid metals. Furthermore, it is crucial not to overlook the associated downtime and maintenance demands that arise when the structural integrity of the metal guiding tubes is compromised. Such incidents not only disrupt operations but also necessitate extensive repairs, thereby affecting the overall efficiency and productivity of the imaging process. Furthermore, it is worth noting that the reconstruction of free trajectories or those significantly deviating from circular paths necessitates the utilization of more complex and computationally intensive reconstruction algorithms. Consequently, it also requires powerful hardware resources.

Proper and secure fixation of the sample is crucial and may even be considered more critical than positioning. It is imperative to ensure that the sample remains stationary throughout the scan to prevent any movement that could at best only degrade the scan quality but at worst cause bodily harm or damage to the system. Therefore, great care must be taken to ensure that the sample is securely fixed in place to guarantee the accuracy and safety of the scan.

3.2 Challenge regarding maximum acceleration voltage and beam hardening

In order to investigate the possibility of an efficient, industrialized, and non-destructive analysis of riveted joints, including the assessment of their energy-dependent radiodensity, a series of CT scans was performed, emphasizing the importance of achieving sufficient penetration through the sample.

For the series of CT Scans two aluminium plates (48 mm $\times$ 110 mm $\times$ 1.5 mm) have been joined with four half-hollow, self-piercing rivets (5.25 mm shaft diam., 8.0 mm head diam., 5.6 mm height) that are made of Boron steel 38B2 and are commonly used in the automotive industry. While the rivet connections examined here are identical to those used in production, the same cannot be said for the structures to be joined and their surrounding environment. Often, the rivet connections are not non-destructively accessible for CT examination. Vehicle pillars come closest to the experimental conditions presented here. Therefore, it is important to emphasize that even with a robot-based CT system (see Sect. 2.4 and Fig. 2), the quality achieved will, with current technology, be inferior compared to the quality obtained in this study. A total of twelve scans with constant power, yet varying voltage and current were performed. Table 1 shows the most important parameters for each scan. All scans were performed with a Phoenix “v|tome|x L300” industrial CT system, which was equipped with a 300 kV micro-focus tube (see the setup in Fig. 3). The detector was of the type “Dynamic 41” [60]. The detectors offers a $100\,\upmu \hbox {m}$ pixel-pitch and uses a CSI-based scintillator. Every scan took approximately 38 min until completion.

Fig. 3

Experimental setup of the twelve-scan-series

Table 1 Fundamental scan parameters of twelve scans with constant power

For accurate measurements pertaining to the quality of riveted joints, it is imperative to produce clear, high contrast CT imaging data. This necessitates achieving high resolution, ensuring good material contrast, and minimizing CT artifacts as effectively as possible. While the sample used in these measurements does not represent all rivet joints in the automotive industry, it serves to demonstrate the universal quality-degrading nature of beam hardening, specifically in the case of steel rivets. It is generally true that high voltages combined with sufficient physical filters (last column in Table 1) can reduce beam hardening artifacts. This trend can be observed in Fig. 4, which depicts three cross-sections through the same riveted joint at different acceleration voltages. The left side of the figure presents a comparison of the cross-sections from just two of the twelve scans, at acceleration voltages of 70 kV and 290 kV, as the artifact-based differences are most pronounced in these instances. The profiles along the dashed lines of all twelve scans are visualized on the right side next to the cross-sections in Fig. 4. For a clear visualization of the progressive reduction of beam hardening artifacts with rising acceleration voltage ($U_{\text {acc}}$), please refer to the right side of Fig. 4. To enhance understanding, it is recommended to examine the profile projections on the left side of the 3D lines. There, the trend of decreasing artifacts is readily apparent, with the scans conducted at 250 kV, 270 kV, and 290 kV demonstrating the lowest levels of artifacts across the series of twelve scans. Therefore, only data acquired with accelerations voltages larger than 250 kV could have been utilized for dimensioning and quality analysis purposes among all the scans taken.

The profiles in Fig. 4 reveal another notable trend: as $U_{\text {acc}}$ increases, there is a discernible decrease in grayscale contrast. This effect is not immediately evident in the cross-sectional images on the left of Fig. 4, but the variations in signal intensity become quite clear when looking at the profile lines on the right. This observation indicates a necessary trade-off in imaging: either one opts for a lower level of beam hardening artifacts, which enhances structural clarity, or one preserves a wider range of grayscale values. The latter is particularly crucial when scanning materials or structures with closely matched radiodensity. Thus, the choice hinges on the imaging priority—whether it is the delineation of fine details, such as cracks or porosities, or the differentiation of materials with subtle density differences.

The manifestation of beam hardening and the influence of $U_{\text {acc}}$ and physical filters can be observed in more detail in Fig. 5: Images A1, B1, C1: The gap between the two aluminium plates is only slightly visible for the 230 kV and 290 kV instances. Images A2, B2, C2: The dark structure in A2 could be mistaken for absent material. Images D1, E1, F1: In order to measure weather the rivet’s head protrudes the top plate both the rivet head and top plate need to be clearly visible. Note the thin aluminium foil that can only be seen at 290 kV. It was used to print a sample identification number on it. Also note, that the same foil can be seen even at low energies in the yz-plane (images G, H, I). This shows the directional dependence of beam hardening artifacts. Images D2, E2, F2: Another critical characteristic of a rivet joint is the remaining sheet thickness at the rivets feet. Note that compared to the xz-plane (images D, E, F) the remaining sheet thickness can be seen clearly even at 70 kV in the yz-plane (images G, H, I). This shows again the directional dependence of beam hardening artifacts. Images G1, H1, I1: G1 and H1 display a dark structure similar to the one observed in A2 when viewed at low energy levels. This feature could be erroneously perceived as missing material. These particular images focus on the crucial point where the two plates connect with the rivet’s shaft. This point is important because it is necessary for measuring the rivet’s undercut, which is an indicator of the joint’s strength. Artifacts appearing at this critical spot can lead to dangerous misinterpretations, with potentially serious consequences if read incorrectly. Images G2, H2, I2: The top of the rivet head as well as the transition from rivet material to plate material can only be seen clearly at higher energies. Both Figs. 4 and 5 illustrate the importance of a high $U_{\text {acc}}$ and strong physical filtering when automotive rivet joints are to be evaluated in a comprehensive and accurate fashion using conventional laboratory-based CT systems.

Fig. 4

Quantitative analysis of voltage dependent beam hardening artifacts. Left: Three, orthogonal cross-sections (xy-, xz-, yz-plane) through the same rivet at two different acceleration voltages (70 kV and 290 kV). Right: Each graph shows the normalized profiles along the dashed lines in the cross-sections across the entire measurement series from Table 1. Both 70 kV and 290 kV profiles are emphasized with orange and blue, respectively. Prominent beam hardening artifacts can be seen in the cross-sections of the 70 kV scan. They manifest through the hazy smearing of the highly absorbing steel rivets. The 290 kV scan is noticeably clearer in that regard. Although rivets have a high degree of symmetry, directional artifacts appear in all three planes. This directional dependency comes from the transmission direction. The grayscale histograms of the images are scaled in order to make them visually cmparable

Fig. 5

Qualitative analysis of voltage dependent beam hardening effects. Direct comparison of selected areas show strong beam hardening artifacts for low accelerations voltages. Each of the three blocks shows a different cross-sectional plane. Within each block, the same cross-section is shown at three different acceleration voltages (70 kV, 230 kV, and 290 kV). It can be seen clearly in the magnified areas that artifacts begin only to disappear almost completely between 230 kV and 290 kV

It should be noted, however, that the severity of artifacts is influenced by factors beyond just the combination of acceleration voltage and physical filters. The kind of X-ray source and the employed detector impact the quality as well. To demonstrate this, two scans were performed: In one scan, from here on referred to as $\varvec \upmu$CT-scan, a conventional micro-focus X-ray tube (XWT-160-CT, X-RAY WorX GmbH, Garbsen, Germany) was used in combination with a photon-counting detector (XC-THOR, Direct Conversion/Varex Imaging AG, Walluf, Germany). The scan parameters are listed in Table 2. The other scan, from here on referred to as SRCT-scan was conducted at beamline P07 operated by Helmholtz-Zentrum Hereon at the Deutsches Elektronen Synchrotron (DESY), Hamburg, Germany. In this experiment, we utilized a specialized detector developed in collaboration between Helmholtz-Zentrum Hereon and the Karlsruhe Institute of Technology (KIT). Upon closer inspection, the detector comprises a scintillator-lens-CMOS assembly. The scintillator is made from a cadmium tungstate (CdWO4) crystal with a thickness of $100\,\upmu \hbox {m}$. The employed lens (POG Präzisionsoptik Gera GmbH, Loebichau, Germany) was used with a 10-fold optical magnification setting. The CMOS sensor (CMV20000 from CMOSIS Imaging Sensors, Antwerpen, Belgium) features 20 megapixels and a pixel size of $6\,\upmu \hbox {m}$. The scan parameters of this scan are listed in Table 3 [45]. It is important to note that, for logistical reasons, a different sample had to be used, therefore rendering a direct comparison with the scan data from Figs. 4 and 5 meaningless. As context for the following discussion it is also important to mention that the synchrotron scan parameters are very different compared to the twelve-scan-series from before. Nevertheless, conducting an objective and comprehensive quality comparison will provide insight into the potential of synchrotron-radiation-based technology and why such a level of data quality cannot be achieved with conventional laboratory-based CT systems. These results serve as a motivator for future innovation and advancement, pushing the boundaries of what is currently possible with CT technology. Both scans, $\upmu$CT-scan and SRCT-scan, are juxtaposed in Fig. 6 ($\upmu$CT-scan left, SRCT-scan right), Fig. 7 ($\upmu$CT-scan top, SRCT-scan bottom), and Fig. 8 ($\upmu$CT-scan top, SRCT-scan bottom).

Table 2 Parameters of the scan that employed a photon-counting detector and a conventional laboratory CT system

Table 3 Parameters of the scan conducted at the synchrotron facility

In the SRCT-scan, a dominant, pillar-like shadow can be observed, which is due to a combination of stitching and intensity dependent scintillator response. Apart from this structure, both scans are nearly artifact-free compared to the twelve-scan-series, presented in Figs. 4 and 5. It is important to note that both $\upmu$CT-scan and SRCT-scan were conducted at well below 250 kV, namely at 160 kV and 103 kV, respectively. Both scans exhibit a quality that is superior to that of the twelve-scan-series. This is in part due to the incorporation of photon-counting detectors and the superior synchrotron radiation beam quality, both of which are exceptionally suitable for reducing beam hardening artifacts. However, the finest structures can only be observed in the SRCT-scan. Here, even the remnants of the rivet’s coating (the white coating and particles), which provides a galvanic separation between the steel rivet and the aluminum sheet to prevent corrosion [44], can be clearly discerned. The primary distinction between a laboratory-based system and a synchrotron radiation source lies in the energy spectrum. In addition to the energy spectrum, resolution is another crucial factor to consider. The synchrotron scan achieved an effective pixel size of $2.66\,\upmu \hbox {m}$ (binned). In order for a conventional laboratory-based CT system to achieve such a resolution, it would be necessary to lower the voltage and current to a level that is no longer efficient, as will be discussed in Sect. 3.3 and illustrated in Fig. 10. The detector for the synchrotron scan had an area of approximately 2 cm by 2 cm while the beam itself only measured approximately 3 mm by 2 mm. This limitation dramatically restricts the possible sample size. Scanning comparatively large objects, like the rivets presented here, therefore requires multiple scans and subsequent stitching, as well as significant storage capacity. Moreover, the evaluation of computationally intensive volumes necessitates powerful computing clusters.

These observations demonstrate that while it is indeed possible to obtain the necessary data quality for accurately evaluating riveted steel joints, it is not currently feasible to achieve this in a truly non-destructive and industrial setting. As a result, the analysis of such automotive samples is limited to individual, finely tailored specimens. It shall also be noted that operating a synchrotron radiation source, with its significant financial implications, and the necessity of often booking scan times months in advance, poses an industrially unfeasible proposition.

Fig. 6

Comparison of $\upmu$CT-scan (left) and SRCT-scan (right). The cross sections correspond to the yz-plane

Fig. 7

Comparison of $\upmu$CT-scan (top) and SRCT-scan (bottom). The cross sections correspond to the xy-plane

Fig. 8

Comparison of $\upmu$CT-scan (top) and SRCT-scan (bottom). The cross sections correspond to another xy-plane

3.3 Interdependency and Current Limitations of Focal Spot and Power Settings for Conventional Laboratory-Based CT-Systems

Several factors play a crucial role in determining the final resolution of the data, setting the baseline before other influences such as the reconstruction algorithm or storage data format come into play. Most commercially available industrial CT systems employ a sophisticated monitoring and regulation mechanism that adapts the focal spot size based on the resulting power, which is the product of the set acceleration voltage and current. It is common to approximate that $P \approx F$,⁷ where P refers to the tube power load at the focal spot diameter F [15]. Monitoring and regulation mechanisms are necessary to prevent the power-density from reaching levels that could damage the anode surface. An exemplary illustration of a post-damaged surface can be seen in Fig. 9. If the anode surface has been subjected to excessive power levels prior to or during a scan, it can lead to a degradation in the quality of the resulting data, which is attributable to unwanted spectral filtering [6] and distortions in the spot’s otherwise uniform geometry. To mitigate the risk of harming the system, operators are typically prohibited from simultaneously controlling both the spot size and power. However, in some cases where the operator does have control over both parameters, the system usually automatically adjusts the respective set of parameters to safe levels in order to prevent any potential damage.

Fig. 9

Surface of a tungsten coated diamond target that has sustained damage as a consequence of improper operation or malfunctioning of the monitoring mechanism in a micro-focus X-ray tube, leading to excessively high power densities

Apart from the focal spot, there are several other significant factors that influence the resolution in computed tomography. These factors include the magnification factor, denoted as M, which is determined by dividing FDD by FOD, as well as the geometry and dimensions of the detector pixels. The focal spot, however, poses the most significant constraint on the maximum level of spatial resolution that a given CT system can achieve [13]. The extent of blurring caused by the focal spot is called penumbra and characterizes the degree of degradation in resolution:

$$\begin{aligned} F < \frac{P}{M-1} \quad&\text {(No spot-induced blurring)}, \end{aligned}$$

(1)

$$\begin{aligned} F > \frac{P}{M-1} \quad&\text {(Spot-induced blurring)}. \end{aligned}$$

(2)

Here, F is the diameter of the focal spot, P denotes the size of the detector pixels, while $M = {FDD}\big /{FOD}$ is the magnification factor. Geometric resolution R, representing the maximum resolution achievable solely through the pixel size, can be mathematically expressed as follows:

$$\begin{aligned} R = f\cdot \frac{P}{M}. \end{aligned}$$

(3)

The factor f can be determined by calculating the diagonal length of a 3 by 3 pixel array. In this case $f=\sqrt{3^2+3^2}\approx 4.25$. This is to make sure, that the desired feature size can be resolved by at least 9 pixels. Otherwise, geometric effects or artifacts could jeopardize the detectability of the target structure (e.g. small cracks or particles). Through the use of the magnification factor M, a correlation between the focal spot size F and the desired geometric resolution R can be established. This correlation enables the determination of the maximum allowable size for the focal spot for a given pixel size P and a magnification M, ensuring that the desired resolution remains uncompromised. The relationship between these factors is illustrated in Fig. 10. The left side of the graph illustrates the required magnification for various detector pixel sizes, presenting how it relates to the desired geometric resolution. On the right side, the graph displays the magnification as a function of the focal spot diameter. By analyzing these visual representations, one can make informed decisions about the appropriate magnification and the acceptable spot size in order to meet specific geometric resolution requirements. An example has been marked with dotted arrows and numerals in boxes: By commencing at (1) and considering a structure of $200\,\upmu \hbox {m}$ that requires to be resolved, and by assuming a detector pixel size of $74.8\, \upmu \hbox {m}$, we arrive at (2) and (3) and the respective minimum magnification $M=3$. As a result, this mandates in turn a spot diameter no greater than $30\,\upmu \hbox {m}$ at (4). Within the figure, the orange area represents focal spot sizes that would compromise the image resolution, resulting in blurriness, given a specific M and P. Conversely, the green area signifies a region where spot-blurriness is absent, once again considering a specific M and P. This visual representation effectively demonstrates the inescapable correlation between unsharpness, detector pixel size, and focal spot diameter, and emphasizes the importance of carefully managing them to achieve optimal image quality. This intricate interplay primarily focuses on determining whether a specific resolution can be attained based purely on geometric considerations. However, our discussion has not yet addressed the following question: assuming that the discovered geometric parameters, particularly the size of the focal spot, enable the desired resolution, is it possible to adjust the voltage and current in such a way that, on the one hand, the object can be adequately transilluminated by X-rays, and on the other hand, enough photons are generated to minimize the scanning time?

It is crucial to acknowledge that the desired size of the focal spot plays a pivotal role in determining the permissible power utilization, as previously elucidated. Consequently, this restriction imposes limitations on either the acceleration voltage or the tube current, subsequently affecting the penetration depth or the number of photons, respectively. Thus, the size of the focal spot significantly influences the extent to which different materials can be penetrated and the speed at which the scanning process can be executed. These interconnected factors underscore the intricate balance required to optimize both resolution and operational efficiency in this context. To provide a comprehensive assessment of the limitations associated with high-resolution analysis of rivet connections using CT, it is essential to have a robust database of contemporary X-ray tube specifications encompassing the entire market. To accomplish this, a trio of data comprising voltage, power, and their corresponding minimum focal spot diameter is required. These insights will enable us to make a conclusive statement regarding the feasibility and effectiveness of the desired analysis and depiction.

Fig. 10

Correlation between geometric resolution R, pixel size P, and focal spot size F, established through geometric magnification M. The dashed and solid lines that are depicted in both graphs, as well as the intermediate gray area, show pixel sizes derived from a rudimentary digital search, thus making no claim of absolute comprehensiveness. Nevertheless, this range contains common and widely used pixel sized of commercially available detectors used in industrial CT systems

In order to gain a well-founded overview of the current NDT X-ray tube market, an extensive review of publicly accessible resources was conducted. This included examining brochures and websites, and when necessary, direct inquiries made to tube manufacturers and system integrators to obtain these data-triplets. The data is a compilation from leading X-ray source manufacturers and/or system integrators, including Thermo Fisher Scientific Inc., Hamamatsu K. K., Waygate Technologies,⁸ ProCon X-Ray GmbH, Carl Zeiss AG, Werth Messtechnik GmbH, SEC Co. Ltd., Comet Yxlon GmbH, Varex Imaging Corp., Viscom AG, VisiConsult X-ray Systems & Solutions GmbH, Nikon Metrology Inc., X-ray Worx GmbH, and Excillum AB. The outcome of the data-triplet review is visualized in Fig. 11. In it, every line connects a maximum acceleration voltage with a resulting power and the associated smallest possible focal spot diameter. It is important to highlight that the visualization is the result of meticulous data curation aimed at presenting only unique data-triplets. This effort is particularly crucial given that some CT-system manufacturers produce their own X-ray tubes, which could lead to distinct triplets, while others may procure their tubes from the same external suppliers as a competitor, raising the possibility of duplicate specifications. Furthermore, even systems from different manufacturers can share identical characteristics, necessitating a thorough review to avoid redundancy in the presented data. Out of respect for confidentiality, the companies’ names are not attributed to the data in Fig. 11.

Although, data from sources with the capability of reaching up to 600 kV has been collected, in the interest of relevance (see Fig. 4 in Sect. 3.2 and the associated discussion) and enhanced clarity of presentation, systems with maximum acceleration voltages exceeding 350 kV were intentionally omitted and not taken into account for further examination.

For a purposeful analysis with minimal artifacts, an industrial X-ray tube should be capable of achieving voltages greater than 250 kV while also maintaining the smallest possible focal spot. To ensure an efficient scanning operation, it is imperative that the tube can sustain a high level of power in a way that $P \gg F$. As laid out in Fig. 11, the collected data reveals that using industrial, laboratory-based X-ray sources, at the time of writing, no system currently achieves the realization of higher powers—through elevated voltages for improved sample penetration and increased currents for shorter scan times—and simultaneously small focal spot sizes for potentially better resolution.

This predicament demands making a choice between two options: reduced scan times with high currents but lower voltage, consequently yielding lower penetration depths; or lengthier scan duration with less intense currents, but higher voltage for penetrating deeper into the target specimen. In both cases, the smallest possible focal spot is sought.

To ensure a comprehensive understanding, let us touch on how to determine focal spot sizes. X-ray tube manufacturers provide specifications for spot sizes that can change with different power settings. It’s crucial for users to check these specifications themselves regularly once the X-ray tube is mounted and functional in its cabinet. This is due to the fact that components of the X-ray tube, such as the anode surface, focusing cup, and coils, are subject to wear and tear over time and usage. This deterioration can alter the shape and size of the spot, deviating from the manufacturer’s original specifications. Standards like ASTM E1165-20 give guidelines on how to measure focal spot sizes. Common methods include using star resolution patterns, along with pinhole or slit masks [5]. These techniques are relatively simple to perform and are important for ensuring that the X-ray system can deliver the sharpness and detail needed for its specific tasks.

Fig. 11

X-ray source characteristics across leading source manufacturers and/or system integrators visualized with a parallel-plot. Only unique voltage–power–spot-size pairings are shown. Additionally, the request of the aforementioned companies to abstain from assigning individual data-triplets to company names was fulfilled. The red, blue, and orange lines, respectively at 300 kV, 320 kV, and 350 kV, are emphasized for the discussion following the results about beam hardening artifacts from Sect. 3.2. Additionally, the line thickness was adjusted in certain cases to avoid complete mutual obscuring of lines. For a fast and high quality scan without blurring, one would hope for tubes with a high max. $\hbox {U}_{\text {acc}}$, a high max. power but at the same time a small min. focal spot diameter

3.4 Compatibility between quality and scan time

In contrast to the other sections focusing on technological and physical challenges, this section provides an economic perspective. Assuming that a CT system can be adjusted to achieve the required resolution and contrast without excessive artifacts (see Sect. 3.2), the question arises whether the scan can also be performed in an acceptable time. The speed of the scan is mainly influenced by the maximum possible power and an adequately short FDD. However, as discussed in Sect. 3.3, X-ray technology presents a dilemma when it comes to balancing a small focal spot and a short FOD or high magnification M (for higher resolution), high acceleration voltage (for better penetration), and high currents (for shorter scan times). In the case of mechanical automotive rivet joints, achieving a small spot, as demonstrated in Sect. 3.2, requires a high voltage to penetrate the joint adequately, while a low current is necessary to prevent blurring of fine structures and potentially damaging the anode. However, a low current also means fewer photons, resulting in a poor signal-to-noise ratio (SNR).

Alleviating the poor SNR caused by low currents necessitates exposing each projection for a significantly longer time. However, this approach conflicts with the economic constraints of many industrial processes, as it would require waiting longer for a result per joint. Rather than extending the image acquisition time, one can also enhance the SNR by employing the technique of pixel binning on the detector. However, this improvement comes with an inevitable trade-off, as it results in a reduction of the image’s resolution. As an example, employing a 2 pixel by 2 pixel binning technique results in a halving of resolution.⁹ However, simultaneously, the signal acquired for each of these newly formed pixels is effectively quadrupled, as this process consolidates the signal from four original pixels into one larger effective pixel.

In the laboratory-based experiments (Sect. 3.2), a resolution of $30\,\upmu \hbox {m}$ was achieved with a 38-minute scan. However, to minimize artifacts, high voltages (above 250 kV) were necessary. Maintaining a small spot also requires relatively low currents. The synchrotron scans in the same section revealed that even higher resolutions might be required, as the contact point between the rivet and the upper and lower plates can be less than $10\,\upmu \hbox {m}$. Achieving such high resolutions would necessitate even smaller spots and lower currents, resulting in longer exposure times. In conjunction with Figs. 10, 11 serves to highlight the significant challenges involved in non-destructively testing riveted joints using CT in a high-resolution manner and within an acceptable time frame.

4 Conclusion

In this work, we embarked on an investigation into the challenges encountered in non-destructive X-ray CT testing of riveted joints in the automotive industry. Our aim was to address four fundamental questions relating to accommodation and safe positioning of the sample structure, radiodensity and beam hardening, the relationship between spot size and power, and the overall economic feasibility of different settings. While the first three questions pose technological and physics-related challenges, the latter primarily pertains to economic considerations. In order to have a comprehensive overview of the current state of industrial X-ray tubes, we also discussed key aspects and highlighted promising technologies and concepts at the appropriate sections. However, we also recognized that these approaches have their respective limitations, which must be considered when implementing them.

In a first experiment, we performed twelve scans and utilized a standard commercial laboratory-based CT system with a micro-focus tube and an energy-integrating FPD. We demonstrated that at acceleration voltages between 70 and 230 kV beam hardening artifacts had a detrimental impact on the quality of CT scans of steel rivet joints, despite physical filtering. Consequently, the utilization of higher voltages (250–290 kV) and appropriate physical filtering became imperative to mitigate these effects. However, even with these measures in place, achieving resolutions comparable to those obtained through macro-sectioning and microscopy proved to be virtually impossible. While the samples used in these measurements do not represent all rivet joints in the automotive industry, they serve to demonstrate the universal quality-degrading nature of beam hardening, specifically in the case of steel rivets.

To gain a deeper understanding of the impact that the characteristics of the X-ray source and the technology of the detector have on the quality of the scans, further examinations were carried out: one at a synchrotron facility and another employing a conventional micro-focus tube paired with a photon-counting detector.

With a beam energy of only 103 kV the combination of synchrotron radiation and the scintillator-lens-CMOS assembly achieved a scan quality that is indeed comparable to that of macro-sectioning and microscopy, which is presently considered the gold standard in the field. Although synchrotron imaging provides superior quality compared to conventional laboratory-based CT systems, it is limited to small, precisely cut and cropped samples (in the case of rivet joints presented in this work, 15 mm by 15 mm). As a result, this approach is not a purely non-destructive method, let alone economically viable. The exceptionally long scan duration of 14.5 h was in part necessary due to the requirement to image the entire sample while being constrained to a very small FOV. The scan time could be significantly reduced, for example by focusing on smaller subregions. Therefore, synchrotron imaging may be well-suited for investigating research inquiries regarding rivet joints and similar mechanical structures. However, it is not suitable for inspecting large quantities typically encountered in in-line or at-line quality inspection processes within industrial environments.

The combination of a conventional micro-focus X-ray tube and a photon-counting detector has also surpassed the previously conducted twelve-scan-series, albeit solely in terms of reducing beam hardening artifacts. Remarkably, this was achieved with an acceleration voltage of only 160 kV. This underscores the significant impact that the choice of detector can have. However, the achieved voxel size of the synchrotron scans, which was $2.5\,\upmu \hbox {m}$, could not be matched. This was due to the respective setup and the vastly different detector pixel sizes; in the case of the synchrotron detector, the pixels were approximately 151 times smaller than those of the laboratory-based photon-counting detector. Furthermore, this configuration also necessitated an exceptionally long scan time (12.1 h) to attain an adequate data quality. In summary, regarding photon-counting detectors, it can be stated that although they are capable of achieving the necessary contrasts with minimal artifacts at comparatively low energies, their economic relevance is diminished due to the prolonged exposure times required—at least when standard laboratory X-ray tubes are employed.

Moreover, our investigation into the current industrial NDT X-ray tube market revealed that the stringent conditions necessary for efficient, non-destructive X-ray CT testing of riveted joints in the automotive industry are not satisfied by any commercially available off-the-shelf CT system. We gathered pertinent data from leading X-ray tube manufacturers and system integrators, and our analysis indicated that none of the systems could concurrently provide the requisite voltage and power levels while preserving a sufficiently small focal spot size, criteria that are essential for an effective and economically feasible industrial application. This aspect is particularly critical in scenarios where photon-counting detectors are not utilized. Our preceding experiments with conventional energy-integrating FPDs have demonstrated this, suggesting that higher acceleration voltages are necessary. However, our experiments employing photon-counting detectors indicate that significantly lower energy levels are sufficient to produce contrast-rich and high-resolution data, challenging the previous assumption that higher voltages are always required for optimal imaging of rivet joints.

In addition, the economic feasibility of routine quality control in manufacturing settings will be significantly impacted by the cost and maintenance of an X-ray CT system and the time required to conduct a scan that eventuates in the desired quality. This further emphasizes the need for continued research and development efforts to overcome the remaining obstacles and enable the widespread implementation of a testing method in an economically viable manner.

5 Outlook

Currently, the laws of physics and technological limitations make it highly improbable to develop an industrial solution that can non-destructively provide an efficient in situ analysis with high-resolution, volumetric data of entire car body structures and their joining connections. Based on our findings, the synergistic integration of a transmission tube, designed to maintain small focal spots, in concert with photon-counting detectors, and robotic arms for improved reachability (variable zooming, free trajectories), may well be the key to unlocking the desired outcome.

Another viable solution might be the LMJ anode. It guarantees a high number of photons with a small focal spot, thereby reducing scan time yet delivering high resolution. Nevertheless, integrating this system onto robotic arms for in situ car body inspections could present challenges, as the abrupt motions of the robots may compromise its stability. Additionally, the setup may encounter difficulties stemming from the dynamics of the liquid metal circulatory system and its rigid high pressure tubes. The size of current LMJ sources would also restrict the reachability of the system.

Given the pivotal role photon-counting detectors play in various industries, it is imperative that future research endeavors shed light on the potential benefits and challenges associated with their implementation in the automotive industry. Such a setup might even enable the truly non-destructive analysis of structures such as car pillars.

For integrating synchrotron-radiation-based investigations as a gold-standard for the shown types of objects an optimized beamline providing high-energy synchrotron-radiation microtomography using a centimeter-sized X-ray beam together with a high throughput and an easy access would be desirable.

Following data acquisition, the manual and often laborious and time-consuming evaluation and dimensioning of rivet joints take place. However, with the significant advancements in machine learning and artificial intelligence in recent decades, the potential for these tasks to be fully automated and efficiently executed by algorithms is within reach. Emerging approaches that promise automated and calibrated measurement of rivet connections on micrographs are already implemented. The success and prevalence of such technologies will be pivotal in enhancing the efficiency and economic viability of NDT methods in industrial environments.