Exploration of machine learning approach to predict ultra-high temperature mechanical properties of carbon/carbon composites from multi-scale microstructure parameters

The evolution mechanism of mechanical properties and the corresponding prediction model constitute a pivotal topic in composites science. Nevertheless, accurately establishing the intricate nonlinear microstructure-performance relationship based on experimental and theoretical approaches remains inherently challenging. In this paper, machine learning (ML) is employed to map the experimentally obtained multi-scale microstructure parameters of graphene, carbon fibre (CF) monofilaments and composites to the ultra-high temperature (up to 2950 °C) mechanical properties of carbon/carbon (C/C) composites. The results demonstrate that the advanced data-mining tools, such as artificial neural network (ANN), can significantly improve the prediction accuracy of the mechanical properties, specifically, the R² scores of the tensile strength and tensile modulus of C/C composites are 0.9973 and 0.9989, respectively. This research further verifies that ML is effective in analysing the influence ranking of characteristic parameters (e.g., the graphite crystallite size and interface strength prove critical for C/C composites design) and exploring the variation mechanism, which facilitates the optimisation of C/C composites process parameters according to the requirements of the practical application and deepen the in-depth understanding of underlying physical phenomena.