ADAPTIVE APPROACH FOR SEGMENTATION OF LOW-CONTRAST µCT IMAGES IN THE CASE OF PARTIALLY IMPREGNATED WEFT-KNITTED COMMINGLED YARN COMPOSITES - École nationale supérieure Mines-Télécom Lille Douai Accéder directement au contenu
Communication Dans Un Congrès Année : 2021

ADAPTIVE APPROACH FOR SEGMENTATION OF LOW-CONTRAST µCT IMAGES IN THE CASE OF PARTIALLY IMPREGNATED WEFT-KNITTED COMMINGLED YARN COMPOSITES

Résumé

The purpose of this work is to provide an objective procedure to conduct qualitative and quantitative analyses based on microcomputed-tomography (μCT) images of different contrast levels which can be obtained in the case of partially impregnated technical preforms of thermoplastic composites. The considered 4D dataset was constructed using absorption-based μCT scans using the same set of parameters to collect X-ray sinograms with a voxel size of 10 μm. The scanned samples represent four specimens of the same grade of a weft-knitted fabric formed of commingled glass fibers and polypropylene which were prepared at four compaction ratios ranging from zero (no compaction) to 0.63 (highest compaction level). Objective image-quality descriptors such as the image-contrast and the signal-to-noise ratio were evaluated [1]. These descriptors were required to shed more light on the combined effect of the mechanical deformation of the architectured fiber-bed and its impregnation by the thermoplastic matrix on the quality of the reconstructed μCT images [2]–[4]. Comparative analyses of these descriptors indicated that beam hardening is the most significant source of reconstruction errors in the case of the non-consolidated composite samples and that partial volume effect makes it difficult to visually distinguish between dry and impregnated zones inside the yarns at a compaction ratio of 63%. The study was then extended to quantify residual porosity with respect to the respective contrast-levels of the used 3D μCT images (where porosity represents interstices between the reinforcement fibers and the yarns which were unfilled by the thermoplastic matrix). A weight-controlled thresholding procedure was suggested to overcome the contrast variability and was then applied to test the reliability of unsupervised automatic segmentation techniques such as the K-means clustering method for porosity segmentation [5]. Obtained results indicated that the highest accuracy levels required an image contrast of about 0.5 and a signal-to-noise ratio lower than three. The conducted work provides insight about the required image quality levels to conduct objective μCT-based analyses of partially impregnated commingled GF/PP weft-knitted fabrics.
Fichier non déposé

Dates et versions

hal-04063135 , version 1 (08-04-2023)

Identifiants

Citer

Abderrahmane Ayadi, Chung Hae Park. ADAPTIVE APPROACH FOR SEGMENTATION OF LOW-CONTRAST µCT IMAGES IN THE CASE OF PARTIALLY IMPREGNATED WEFT-KNITTED COMMINGLED YARN COMPOSITES. 10th International Conference on Composite Testing and Model Identification, May 2021, Lille (FR), France. ⟨10.13140/RG.2.2.11027.81448⟩. ⟨hal-04063135⟩
6 Consultations
2 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More