Senior manager, Mechanical Design and Analysis, Rolls-Royce Inc., Atlanta, GA
Assistant professor, Auburn University, Auburn, AL
In this paper, a novel visual quality control technique for production of textile composite materials is presented. This technique is based on the idea of automated image processing. The approach starts by acquiring images of the surface of a composite product using a combination of a microscope and a frame grabbing board connected to a computer. Image processing is then applied to the acquired images to extract important features. In the current study, the features of importance are orientation of surface yarns.
The image analysis approach is constructed taking into consideration the intrinsic characteristics of the acquired images. Important image features are enhanced using a Difference of Gaussians and a set of directional edge detection kernels. The enhanced image is then thresholded using a “Fixed Percent Setting” technique and converted into binary format. Hough Transform and Bounding Box approaches are utilized for object (that is, fibers) recognition in the binarized image.
This approach is carried out on 20 ceramic composite parts. The fabric preform was manually placed in all these parts. The technique was successful in determining predominate directions orientation of surface yarns in most of the parts. Matrix material in some areas of these parts were over-grown in the infiltration process. Due to this phenomena, 6% of the acquired images have no data available about yarn's orientation. Consequently, the image analysis approach is unsuccessful in obtaining information about yarn directions in these areas.
Paper ID: CTR10460J