Abstract
A curated multimodal microscopy dataset is presented for statistical quantification of through-thickness, bundle-scale impregnation and porosity-defined as non-filled regions within bundle cross-sections-in woven glass-fibre/polypropylene (GF/PP) laminates. The dataset spans three controlled compaction ratios (Cr = 0 %, 30 %, 41 %), providing a structured basis for investigating processing-microstructure relationships in thermoplastic composites and addressing limitations of 2D single-modality analyses of partially impregnated materials. Large-area cross-sections were prepared post-processing using a fluorescence-enriched epoxy mount and a multi-step polishing protocol tailored to partially impregnated thermoplastics, then imaged by fluorescence microscopy (FM), polarized light microscopy (PLM) and backscattered-electron scanning electron microscopy (SEM-BSE). Overlapping tiles were stitched into millimetre-scale extended-field images that still resolve individual filaments and were rigidly registered to form aligned FM/PLM/SEM stacks across the full laminate thickness. Detailed procedures for sample manufacturing, surface preparation, polishing and quantitative analysis are provided in a companion research article by the same authors [1]. The core quantitative products are extended-field FM and SEM images with associated five-class Random Forest segmentation maps, trained on image-derived intensity features to distinguish phase- and porosity-related classes (including glass fibres, matrix and void-rich regions), while PLM primarily documents surface state and polishing quality. For bundle-scale analysis, >15 complete fibre bundles oriented perpendicularly to the polishing plane (0°-oriented) are extracted and systematically labelled per extended-filed image. For each bundle, the dataset provides paired FM/SEM crops, corresponding segmented images, and binary masks for void-limited masks, bundle outlines and the glass-fibre phase. The companion research article reports per-bundle quantitative metrics, whereas the present data paper describes the dataset structure and known limitations, including large image sizes, residual SEM brightness drift, exclusion of bundles with pronounced stitching artefacts, and the non-linear response of dye infiltration to porosity. The full dataset, including raw and processed image products, is available in the public repository Recherche Data Gouv [2]. To the best of current knowledge, this is the first publicly available FM/SEM multimodal dataset at single-fibre resolution over millimetre-scale fields of view for thermoplastic composite microstructures, providing a benchmark resource for registration, segmentation and impregnation-quantification methods.