UC Berkeley

Fiber-reinforced polymer (FRP) composites are widely used in aerospace, marine, automotive, and other industries due to their superior strength-to-weight ratio and corrosion resistance. Analyzing FRP cross-sectional micrographs is one of the most widely used approaches for defect detection, quality inspection, failure analysis, and computational materials modeling. Although micrographs are widely used, a lack of specialized image and geometric processing techniques forces materials science researchers to manually analyze images, which makes the analysis process time-consuming and error-prone.

In this research, efficient and scalable geometric processing algorithms are proposed in order to characterize FRP materials and inspect their microstructure from microscope images. We develop two methods to automatically identify a major defect as well as microstructural feature in FRP composites: the resin-rich areas, which refer to areas of reduced strength caused by the lack of fiber reinforcements. We apply the concept of alpha-shapes and alpha-hulls to formalize mathematical definitions of the boundaries of resin-rich areas, and design efficient and scalable algorithms to compute the defined boundaries. In addition, a fiber recognition algorithm that automatically identifies and evaluates the breakage of the fiber cross-sections, and a GPU-based algorithm that efficiently constructs Voronoi diagrams of spheres/circles, are designed.

These methods enable us to provide statistical analyses to quantitatively characterize the identified resin-rich areas. The rigorous mathematical definition of resin-rich areas and ability to collect thorough statistics will facilitate better understanding and quantification of the relationship between resin-rich areas and material properties.