Identification and evaluation of optimal fiber shape for transverse thermal conductivity of unidirectional composites

As a constituent of composites, fibers play a pivotal role in composites’ properties. In this study, an identification process for fiber shape is developed to enhance the transverse coefficient of thermal conductivity. The identification process is established by combining feedforward neural network, Kriging surrogate model and NSGA-II algorithm. The identification process leads to the discovery of an ellipse-like fiber shape which results in a 202.7% increase in thermal conductivity. However, the equivalent thermal conductivities of the models with ellipse-like fiber are highly dependent on fiber shape orientation. Another fiber shape, gear shape, is identified and a more than 100% increase in thermal conductivities is achieved along different fiber shape orientations. The effects of the fiber random distribution patterns are revealed with different fiber shapes considered. The results demonstrate that when the fiber random distribution pattern is considered, the thermal conductivity of model with gear fibers exhibits the largest increase due to the effects of fiber shape orientation. Furthermore, the equivalent transverse thermal conductivity can be enhanced by the non-circular fiber shapes only if the fiber-matrix thermal conductivity ratio is larger than a certain value which is dependent on the fiber shape. With the increase of the fiber volume fraction, the optimal fiber shape becomes more and more similar to the circular shape and the effects of non-circular fiber shape become less.