Multi-scale correlation of impact-induced defects in carbon fiber composites using X-ray scattering and machine learning

www.nature.com/scientificreports OPEN Multi-scale correlation of impactinduced defects in carbon fiber composites using X-ray scattering and machine learning Alexander H. Sexton1,2, Heikki Suhonen3, Mathias K. Huss-Hansen4, Hanna Demchenko1, Jakob Kjelstrup-Hansen4,5, Matthias Schwartzkopf6 & Matti Knaapila1,2 Impact-induced defects in carbon fiber-reinforced polymers (CFRPs)-spanning from nanometer to macroscopic length scales-can be monitored using an aggregate of X-ray-based methods, but this is impractical in typical field conditions. We report on a low-velocity impacted CFRP, which is mapped using small- and wide-angle X-ray scattering and X-ray computed tomography, and employ machine learning for correlating material parameterizations derived from these techniques. The observed 1 µm to 1 mm-sized defects are parameterized in terms of relative density and fiber orientation indicative of fiber failures (kink bands), and the nanometer sized defects in terms of crystal size and unit cell frustration. The 30 to 300 nm defects are parameterized by a power-law scattering decay, differentiating fractal-like behaviors. We find three spatial domains experimentally and by K-means Clustering: Domains of severe damage (with a visual dent), intact domains (without visual or measurable defects) and a transition domain (defects measurable by X-rays). How the parameters are correlated and how they overlap between the domains are discussed. All parameters are able to point to the detrimental fiber breakage in the severe damage domain, and scattering decay also in the transition domain, for example. How individual parameters determined from one experimental technique can be predicted from that of another is also described. Continuous carbon fiber-reinforced polymers (CFRPs) consist of macroscopically aligned fibers within thermoplastic polymer matrices. Due to their high durability and strength-to-weight ratio, these materials are in increasingly high demand for applications within aerospace and automotive, pressure vessels, wind turbines, protective armors as well as for high-end sporting equipment1,2. Though, during materials processing or operation, the composites may become damaged, either by overloading, fatigue or various impact events, leading to defect types such as matrix cracking, delamination and fiber failure, which might be detrimental to the integrity of the material3–6. These defect hierarchies range from nanometer to macroscopic length scales7, and several experimental methods are therefore required for obtaining a complete picture of the material state. Structural health monitoring (SHM)8 can be conducted using methods which include ultrasonic scans9, acoustic emissions10, vibrothermography11, surface reflectometry12, infrared thermography13, terahertz imaging14, microwaves15, and dielectric response16. Typical for these efforts is to seek for complicated relations which ultimately would allow for accurate lifetime prediction of real-world composites of significant structural complexity, and not only model materials. Consequently, this may lead to elaborate datasets and time-consuming analysis, but defect monitoring should at the same time allow for fast and economical quality control in production lines and field conditions etc. X-ray-based methods are ubiquitous in studies of polymer materials, composites and their defects. For examining defects in composites, much emphasis has been put on X-ray computed tomography (CT), which includes quantifying matrix cracking and inter-ply delamination during shear deformation17, studying crackinitiation processes18, characterizing void and fiber distribution in 3D-printed composites19, predicting void nucleation20, and for investigations of defects as a result of compression loads21. In describing defects and mechanisms at the molecular and intermolecular levels, small-angle and wide-angle X-ray scattering (SAXS/ WAXS) have been employed for investigations of interface structures22,23 and fiber orientation24,25. Attention 1Department of Physics, Norwegian University of Science and Technology, 7491 Trondheim, Norway. 2DPI , P.O. Box 902, 5600 AX Eindhoven, The Netherlands. 3Department of Physics, University of Helsinki, Helsinki, Finland. 4Mads Clausen Institute, NanoSYD, University of Southern Denmark, 6400 Sønderborg, Denmark. 5SDU Climate Cluster, University of Southern Denmark, 5230 Odense, Denmark. 6Deutsches Elektronen-Synchrotron DESY, 22607 Hamburg, Germany. email: ; Scientific Reports | (2024) 14:24393 | https://doi.org/10.1038/s41598-024-76105-6 1 www.nature.com/scientificreports/ has been placed on mapping crystallinity during tensile and fatigue testing26,27, as well as on monitoring microcavities28,29, unit cell frustration and lattice strains30,31. However, the literature directly connecting characterizations across relevant length scales is for carbon fiber composites less comprehensive, even though it has been attempted in other composite materials32. Such characterizations by X-rays do not only depend on several instruments, but also large datasets, which means another problem for fast quality control. Machine learning (ML) methods are commonly employed to find correlations in large amounts of data for SHM of composites and polymers33,34. These methods include Support Vector Machines and K-means clustering combined with mechanical measurements35, acoustic emissions36, dielectric measurements37, electron microscopy38, to mention a few. Where large datasets are obtained from synchrotron radiation, ML methods are used in pre-processing and real-time analysis39. For example, deep learning techniques have been used for visualizing 2D SAXS patterns in low-dimensional latent spaces, allowing for rapidly capturing key features and trends in large datasets, as well as a priori exploration of processing parameter spaces40. Similar methods are also in use for classifying crystallographic dimensionality and space-groups from diffraction patterns41, and for classifying SAXS data in terms of appropriate models for analysis, alleviating the need for domain experts42. When the combination of X-rays and ML in SHM is concerned, segmentation of tomographic data has been successful43, predicting tomograms from data of other SHM methods44, as well as using crystallographic data as predictors for defects45. We have recently reported similar ideas for continuous CFRPs and demonstrated applicability of WAXS together with ML46. Interestingly, we have been able to classify early stage tensile defects that show minor effect that are fully invisible for any macroscopic or visual defect monitoring. This encourages us to proceed with a full X-ray characterization, and defects reaching from early stage to severe stage and from nanoscale to macroscopic length scale. In this paper, we study a low-velocity impact-tested continuous CFRP with polyamide-4,10 with visual damage. We begin with CT and connect this to a comprehensive SA (...truncated)