Investigating Process-Microstructure-Property Relationships and Degradation in Semi-Crystalline Thermoplastic Composites through Multi-Faceted Experiments and Machine Learning Techniques

Abstract

Changes in processing affect the morphology of semicrystalline thermoplastic composites, and, in turn, their performance, including tensile strength, toughness, and high strain-rate sensitive properties such as impact performance. Additionally, thermal degradation or partial oxidation of thermoplastic composites may severely affect the crystal growth process and mechanical properties as well as chemical resistance to common solvents. This also influences the recyclability of the material, as well as available repair-time or time to bring large-scale parts to melt. However, the process-microstructure-property relationships and degradation of thermoplastic composites are not well understood, limiting their applications in industries such as aerospace and defense. In this study, a novel framework of multi-faceted experiments and Machine Learning (ML) Techniques is developed to address limitations of traditional approaches and accelerate the establishment of these relationships in thermoplastic composites. The evolution of crystalline morphology in carbon fiber reinforced thermoplastic composites, such as reinforced PEEK, PEKK, and LM-PAEK, is studied in-situ using a polarized microscope equipped with a heating and cooling controlled stage, combined with probabilistic ML analysis using Gaussian Process Regression (GPR). First, ultra-thin films of low volume fraction composites are prepared and processed with various temperature cycles under the microscope. A classification ML model based on the Yolov3 algorithm is developed to determine size and growth rate statistics, while a segmentation model based on the U-Net algorithm is used to segment crystals that have impinged and do not have clear edges. Next, the size and frequency of transcrystalline and spherulite regions are correlated to temperature history using a combination of traditional theory-based methods for fitting crystallization kinetics, as well as the GPR method. While spherulite growth rate is found to obey the established Lauritzen-Hoffman theory, transcrystalline growth deviates from the theory. This difference is investigated by deconvoluting the underlying mechanisms of growth and impinging of spherulites on transcrystalline regions around fibers. Degradation effects are investigated using a combination of polarizing light microscopy (PLM), Fourier transform infrared (FTIR) spectroscopy, x-ray scattering, and ML techniques including Principal Component Analysis (PCA) and GPR. While PLM is used for in-situ investigation of the effect of degradation on crystallization, FTIR is used for in-vitro analysis of degradation effects on chemical signature of the material. X-ray scattering is employed to measure nanoscale crystalline structures. ML methods are utilized to investigate and map the underlying correlations in high-dimensional, sparse, and noisy experimental data. Results show a significant correlation between melt temperature and processing environment with the resulting lamellar thickness and overall degree of crystallinity. PEEK film, single ply, and 2-ply PEKK prepreg coupons are tensile tested to failure in a Linkam MFS350 at room temperature and 160°C. Additionally, the effect of strain rate on tensile strength is evaluated. Results are used to investigate the effect of polymer degradation (induced by elevated melt temperature, processing in air, and repeated cycling) on tensile strength. The investigation reveals that an increase in strain rate leads to higher tensile strength, whereas undergoing cycles reduces the tensile strength of PEEK film. In the case of 1-ply prepreg, its tensile strength remains largely unaffected by various processing parameters, with the notable exception of test temperature. For 2-ply prepreg, the combined effects of cycling and environmental exposure are significant; specifically, subjecting the 2-ply prepreg to three cycles in an air environment results in a reduction of its tensile strength. The degree of crystallinity in 2-ply samples is characterized using WAXS and DSC and is found to not correlate with melt, cycles, or environment. The fractured surfaces are imaged using SEM. It is determined that the fiber interface may be affected by thermal degradation even when the degree of crystallinity is unaffected. Laser-assisted automated fiber placement is employed to fabricate wedge peel coupons of thermoplastic composites with different processing parameters, including compaction force, roller head speed, and temperature. Using a combination of wedge peel testing to measure the peel force of prepreg tapes, digital and scanning electron microscopic imaging, image segmentation of fractured surfaces, and probabilistic surrogate machine learning fitting, the effect of processing parameters on mechanical performance and failure is examined. Wedge peel fracture surfaces are photographed and analyzed using Weka image segmentation to identify regions of bonding and non-bonding. Scanning electron microscopy is used to study the failure mechanism. These results are then used to train surrogate probabilistic machine learning models to analyze the effect of processing parameters on wedge peel failure. It is observed that the bonding surface area increases with processing temperature, resulting in an increase in wedge peel force. However, increasing the compaction force results in greater variability of the bonding and peel force, while roller head speed has a negligible effect for the examined range. It was noted that wedge peel samples with increased variability tend to crack during testing, exhibiting partial resin and fiber pull-out. The combined image analysis, mechanical testing and machine learning approach successfully characterize the wedge peel behavior based on changes in automated fiber placement processing parameters. A GPR framework was successfully applied to the degradation parameters and correlated with tensile strength. With additional data this framework could be used to achieve a deeper understanding of the mechanics behind thermoplastic degradation in composites.