Phase-contrast and dark-field imaging for the inspection of resin-rich areas and fiber orientation in non-crimp vacuum infusion carbon-fiber-reinforced polymers

The high strength at moderate weight in combination with superior corrosion and fatigue properties [1, 2] makes carbon-fiber-reinforced polymer (CFRP) an attractive material for lightweight applications in aerospace. Nonetheless, besides several benefits, CFRP components also bear significant risks like a low resistance to impact damage [3]. Additionally, CFRP parts are often characterized by the presence of deeply enclosed porosity, resin-rich areas and deviations from the desired fiber layup. These internal features and part-to-part variations can pose a serious threat to the mechanical integrity of a component. Therefore, particularly for safety critical parts used in aviation and aerospace applications, fast and reliable non-destructive investigation methods are essential. X-ray microcomputed tomography (XCT) is a widely established method for damage detection and characterization in CFRP [4, 5] but also other techniques such as ultrasonic testing [6‐8] and thermography [9, 10] have proven suitable for the detection of internal pores and defects. In contrast, the segmentation of resin-rich areas and carbon fiber bundles is more challenging as contrast between fibers and matrix material (resin) is usually insufficient using XCT. Although, at sufficiently high resolution (often less than 1 µm³ voxel size), individual fibers can be segmented and fiber orientation can be determined [5, 11]. However, carbon fibers are too small to be visualized at reasonable specimen dimensions, making the combined investigation with resin-rich areas inefficient, as these are typically a multitude larger in size.

The performance of CFRP laminates is predominantly determined by their internal composition of fibers and matrix material. In particular, the orientation of carbon fibers with respect to occurring stress has a major impact on service life. Depending on load direction, even minor deviations from an ideal fiber orientation can introduce shear stress to a component´s matrix material, significantly weakening its mechanical strength [11]. Such deviations are for example commonly introduced by fiber waviness, and their negative effects on stiffness and strength have already been explained in various studies [12, 13]. Additionally, the stacking sequence of fiber layers influences mechanical properties. Fuoss et al. [14] found that stacking of adjacent layers in similar orientations should be avoided. Consequently, the prediction of mechanical behavior of CFRP components largely depends on a detailed knowledge of aforementioned deviations from an intended/ideal composition. However, Barile et al. [15] reported different levels of compressive stiffness in symmetric and balanced carbon fiber composites depending on compression direction. This suggests that apart from fiber orientation, process parameters such as resin flow direction and local accumulations of resin might as well influence mechanical properties of CFRP.

Depending on the type of composite, resin-rich areas can be caused by multiple reasons, for example, by binder yarns in woven composites that are collimating carbon fibers and consequently creating regions with low fiber content [16]. Similar effects have been observed during the integration of structures significantly thicker than carbon fibers causing a distortion in the composite layup such as embedded electronics [17], optical fibers [18] or resin vasculature in self-healing CFRP [19]. The collimation of fibers or general disturbances of the composite layup and the resulting lack of fibers in those areas can furthermore lead to an entrapment of air during resin infusion creating void spaces in the composite. The resulting porosity can significantly reduce the mechanical strength of a component, e.g., inter-laminar shear strength decreases linearly by roughly 6% for every increasing percent of porosity [20]. However, resin-rich areas in components manufactured by vacuum infusion also occur independently of integrated structures, predominantly in regions where fibers are strongly bent causing a compression of fibers and leaving gaps between layers that are filled with resin during the production process [21]. The major problem of these resin-rich areas is the usual mismatch in stiffness and thermal expansion/contraction in comparison with carbon fibers [21, 22].

During the resin infusion process, dry fiber sheets are placed in an airtight mold and resin is infused by underpressure impregnating the carbon fibers [23]. The subsequent curing process typically involves higher temperatures followed by a cooling of the composite during which internal stress is induced [19, 21]. In addition, most resins have low thermal stability leading to a decomposition of the matrix material during occasional lightning strikes, which can create damage hot spots in resin-rich areas [2, 24]. Previous studies also observed that resin zones from binder yarn collimation correlate with crack initiation sites [25‐27]. For that reasons, the formation of resin-rich areas and their impact on component strength is equally important to investigate.

Using Talbot–Lau grating interferometer (TLGI) X-ray microcomputed tomography (XCT), this study explores three different X-ray contrast mechanisms to characterize internal features and defects in vacuum-infused aerospace-grade CFRP parts. TLGI-XCT is a novel method for the investigation of CFRP and aforementioned problems that was originally developed for synchrotron facilities in 2002 [28]. Because of the high costs and subsequent low accessibility of such facilities, efforts for the transfer of the technology to laboratory computed tomography devices were made. By use of an additional source grating, the obstacle of insufficient spatial coherence of typical laboratory X-ray systems was successfully overcome [29, 30] which opened TLGI-XCT for a wider range of applications. In addition to standard absorption contrast (AC), TLGI-XCT provides two complementary, simultaneously measured and mutually aligned imaging modalities, differential phase contrast (DPC) and dark-field contrast (DFC). DPC shows high sensitivity to changes in atomic number [31] and is therefore the ideal modality for the visualization of low-density material compositions such as biological tissue [32, 33] and the reduction of metal artifacts [34, 35]. On the other hand, DFC visualizes scattering of X-rays, e.g., caused by small structures and defects like cracks or pores in a component. Since the DFC signal provides structural information on a sub-pixel scale, it is possible to quantify defects in larger specimens at relatively low resolution. Moreover, by using one-directional gratings in TLGI-XCT, the sensitivity of DFC is limited to the direction perpendicular to the grating orientation and can therefore be used for the determination of spatial orientation in microstructures [36, 37].

Current research is largely focused on the application of TLGI-XCT for medical diagnosis of chronic obstructive pulmonary diseases [38, 39] and breast cancer [40, 41] but also industrial applications such as the visualization of micro-cracks [42‐45] and fiber orientation in (short-) CFRP [46‐49]. For that purpose, continuous developments regarding the optimization of gratings [50, 51], interferometer setup [52, 53] and reconstruction methods [54, 55] as well as image processing and visualization [56, 57] are still necessary in order to tackle the limitations of the technology [58].

For completeness, it should be mentioned, that there are several more methods for the extraction of DPC and DFC data explained in the literature. Endrizzi [59] provides a comprehensive overview of phase-contrast imaging techniques including analyzer crystal, propagation, edge illumination and tracking-based methods. Similarly to TLGI-XCT, these methods have shown to be suitable for biomedical and industrial applications such as mammography [60, 61], inspection of electronic components [62] and impact damage in CFRP [63].

In this work, we investigated vacuum infusion CFRP specimens extracted from larger airplane components, focusing on curved regions that cause a bending of the internal carbon fibers and a disruption of the internal meso-structure. Since TLGI-XCT is capable of detecting scattering structures in the sub-pixel regime [64], requirements in spatial resolution for the identification of carbon fibers are considerably reduced. Therefore, carbon fibers as well as resin-rich areas in relatively large components can be characterized at once [65] and additional insights into resin-rich areas in combination with local carbon fiber layup and orientation can be found. These insights provide additional, more detailed input for finite elements simulations and consequently enable more reliable predictions for service life. Furthermore, with the implementation of presented methods to product quality control, the additional information can be used for optimization of CFRP production processes and increase cost efficiency.

The capability of TLGI-XCT to detect scattering structures in the sub-pixel regime [64] significantly reduces the required spatial resolution for identification of carbon fiber layers. Moreover, combining the directional sensitivity of DFC and knowledge of the orientation of a visible single fiber layer located at the specimen surface, composite layup, individual fiber layers and their orientation can be readily extracted. Therefore, layers oriented in the same direction as this visually identifiable surface layer were determined by scanning the specimen in the corresponding position with respect to grating direction. The orientations of the remaining layers and the respective specimen placements were subsequently concluded using the information gained from the preceding scans. In that way, the layer sequence was determined with 4 scan positions. This methodology is applicable for standard laminate layups where at least the orientation of an initial layer is known. If this is not the case, prior investigations to evaluate the approximate fiber layer orientations are necessary. This can for example be done by TLGI radiography, as fiber bundles aligned parallel to the optical axis will show a distinct peak in the scatter signal [49]. An approximate assessment of the fiber layup can therefore be done in relatively short time, as radiography images are typically available in live view. However, strongly bent layers can be difficult to visualize because of changes in DFC intensity due to the non-uniform fiber directions. Therefore, in some cases, additional scans are necessary to cover regions with lack of information as visible in Fig. 3 at ± 45° orientations. The increased amount of scans necessary for a single specimen in combination with the typically long scan times caused by the intensity reduction from absorption gratings and phase stepping approach makes TLGI generally rather time-consuming. Additionally, for sake of image quality and constant measurement parameters over all scans performed the scan duration chosen was potentially exaggerated. At the given specimen dimensions, we experienced that a reduction to 375 projection images increases image noise but layer identification by DFC is still absolutely possible. We thereby achieved scan times below 170 min per orientation. Furthermore, for orientations with low transmission lengths (all except 90° orientation) projections or image averaging could be reduced further to achieve scan times comparable to standard XCT. In addition, edge-preserving image filters or iterative reconstruction methods can be applied to counteract increasing image noise at reduced scan times if necessary. Moreover, the SkyScan 1294 system is rather limited in applicable voltage (maximum 50 kV) as its absorption gratings of 40 μm height are too transmissive for higher energy photons. Current state-of-the-art gold gratings can be manufactured with heights of more than 120 μm and curvature fitted to the cone-beam geometry, increasing both the applicable energies and field of view.

For the investigation of resin-rich areas, a single scan is typically sufficient, although the reduction of projections and/or averaging does affect the detectability of resin-rich areas in DPC more severely. In combination with fiber layup identification from DFC data, further conclusions about the cause for certain resin-rich areas are possible. Dong [21] experimentally observed the cause for resin-rich areas between fibers and mold via microscopy in regions characterized by significantly bent layers, resulting in a collimation of fibers. We were able to confirm these observations at regions of interest in an actual aerospace component (see Fig. 7a, b). Furthermore, in [21] only simple bending scenarios of 90° and 135° were performed to provoke the formation of resin zones between fibers and mold. We also found that at more complex bending scenarios, gaps filling with resin can form between individual fiber layers as well. This is particularly obvious between two adjacent layers of 90° orientation, which are characterized by large resin pockets.

However, also binder yarn collimation sites cause resin pockets in composites [16, 26, 72]. As shown in Fig. 7b, these pockets can penetrate through several layers of carbon fibers as they are overlapping each other. Consequently, the formation of such penetrating resin zones is already determined by the placement of fiber sheets in the mold and could possibly be avoided by a shifted arrangement. In extreme cases, the cavities formed by disturbances in the layer sequence can also lead to an insufficient infusion with resin, leaving voids in the composite volume. Since inter-layer resin-rich areas are more pronounced between layers of 90° orientation, we did further investigations revealing that the amount of void volume located between 90° or their adjacent layers is also increased by roughly 20% compared to a random distribution over all inter-layer spaces. Voids in other layers are induced mainly by binder yarn collimation sites.

The additional insights into CFRP laminates gained by TLGI-XCT provide extended structural information for virtual models on a mesoscale facilitating finite element (FE) analyses. Predictions of material behavior of complex CFRP parts necessitate reliable FE simulations that largely depend on accurate experimental data from mechanical tests of physical samples. Detailed knowledge about resin-rich areas and their three-dimensional distribution in composites as well as fiber layup and orientation can therefore contribute to elaborate models for finite elements simulations [25, 27]. Consequently, future works will focus on the implementation of these findings into finite elements models for prediction of mechanical strength and damage initiation. Furthermore, increased scan time of TLGI-XCT in comparison with standard XCT reduces the applicability of the method for quality control in industry. Ongoing research therefore aims a reduction of scan time, e.g., by use of iterative reconstruction methods to reduce the number of necessary projection images.

In contrast to previous studies, this work focuses on the characterization of resin-rich areas and voids in vacuum infusion CFRP through combined information gained by additional image modalities from TLGI-XCT. We have successfully utilized the directional sensitivity of DFC imaging for the determination of an unknown layer sequence. Thereby, the low requirements in spatial resolution enable the characterization of a relatively large volume element in comparison with standard XCT. Furthermore, the high sensitivity of DPC imaging to differences in atomic number was used for the investigation of resin-rich areas in specimens measured. Subsequently, we showed that the information gained from these complementary image modalities can be used to differentiate between in- and inter-layer resin-rich areas and to make more precise conclusions about their cause. Hence, we found that inter-layer resin-rich areas are more pronounced in regions where fiber layers are strongly bent, especially between layers with fibers oriented in bending direction. Other than that, resin-rich areas are mostly caused by the collimation of fiber tows by binder yarn. These findings also correlate with the development of voids in the specimens.