Mesoscale characterization of continuous fiber reinforced composites through machine learning: Fiber chirality

Abstract A method of quantifying fiber chirality, the twist of continuous fibers through a volume, is defined and applied to both phantom data and real data. Specifically, a field quantity termed the fiber chirality based on the anti-symmetric part of the gradient of the fiber orientation is introduced. For this method of estimation, the input is the set of fiber positions gathered from the stack of images which represent the sample volumes. The phantom sample is generated and several different real continuous fiber reinforced matrix composites are experimentally characterized. Each phantom dataset contains a bundle of fibers that rotate about a center axis with a user-defined angle at each step in the z-direction. The chirality of the fibers is calculated based on the pre-characterized positions using a machine learning algorithm. To validate the method of quantification, our chirality estimation method results in a colormap with an angle of rotation that becomes increasingly more similar to the user-defined angle with decreasing inter-slice distance. Fiber positions from real data are then input into the estimation method and the results are compared.