As a consequence of air pollution deterioration worldwide, the global filtration industry is constantly searching for cost-effective solutions with extremely high separation efficiency and remarkably high air flow at low energy consumption for various applications. To answer these market needs, SABIC is developing new nonwoven bi-component spunlaced filer media technology. Studies indicate that an increase in splitting ratio of bi-component fibers enhances the absorption and filtration efficiency, and the mechanical properties of the nonwoven fabric [1]. Currently, fiber splitting ratio is characterized by visual inspection of the Scanning Electron Microscopy (SEM) images, or determined indirectly by measuring filtration properties. There is little information available in literature for quantifying the splitting ratio via image analysis [2].

A novel method has been developed to quantitatively characterize the splitting ratio of bi-component fibers based on SEM images. The workflow for the proposed method is given in Figure 1. As there is no intensity differences between the split and un-split fibers, the thresholding methods for image segmentation are not applicable for this purpose. Alternatively, texture information offers a description of spatially extended patterns of intensity distributions within a neighborhood. The split fibers have smaller diameters, which makes the frequency of edges high. Besides, the changes in orientation are also very frequent because of the entanglement of the twisted fibers. Both texture patterns can be captured by Log-Gabor filter bank [3]. The extracted Gabor texture features are then treated as input to Expectation-Maximization clustering (EM-clustering) to discriminate the two different texture patterns from the two types of fibers. Originally, EM-Clustering is unsupervised machine learning method, which assumes there is a mixture of a definite number of Gaussians within a set of unlabeled data [3]. However, fibers entanglement and fibers depth-penetration make the variations in texture patterns too wide to converge the un-supervised EM-clustering. As such, semi-supervised EM-clustering is implemented instead, which includes a training step on a small set of labeled data to gain some prior knowledge on the targeted Gaussian Mixtures Model. For each image, around 10% of the pixels are randomly selected for training. Initiated with the trained model, pixels belonging to split and un-split fibers are classified further with EM-clustering on extracted Gabor texture features [3]. Post morphology processing follows to finalize the image segmentation. Based on the final image segmentation result, the fiber splitting ratio is then characterized in terms of area percentage of split fiber within the imaged web. Examples of stepwise image analysis results are illustrated in Figure 2. Furthermore, a correlation is established between the derived splitting rate and the permeability of the products. The results on the fiber splitting ratio give a proof of concept in understanding the correlation between filtration efficiency and splitting ratio. This method can also work with SEM images of the sample cross sections to gain an insight of the splitting ratio through the thickness of the sample.

[1]. Baker, B., Bicomponent fibers: A personal perspective, International Fiber Journal, 1998, 13(3), 26-35.

[2]. L. B. Suragani Venu, E. Shim, N. Anantharamaiah and B. Pourdeyhimi, 3D Structural Characterization of Nonwoven Fabrics. Microscopy and Microanalysis, 2012, 8, 1368-1379.

[3]. F. van der Heijden, R. Duin, D. de Ridder, D. M. J. Tax, Classification, Parameter Estimation and State Estimation: An Engineering Approach Using MATLAB, Wiley, 2004.

Figures:

« Back to The 16th European Microscopy Congress 2016

EMC Abstracts - https://emc-proceedings.com/abstract/characterization-of-the-splitting-ratio-of-bi-component-spunlaced-fabric-via-pixel-classification/