Data-Driven Approach to the Prediction of Mechanical Properties in Carbon Fiber Reinforced Composites

For decades, fiber-reinforced composite materials have been integral to the aerospace, automotive, and military industries due to their lightweight properties. The fiber-reinforced composite manufacturing process involves the curing of the matrix material, which is typically a resin, polymer, or ceramic, interlaced with bundles of carbon fibers to form a single lamina; these lamina are then bound to form the laminate. The curing process is dependent on several factors, including humidity, temperature, and cycle time, altogether referred to as the curing environment. Curing environments and cycles are known to have a significant impact on the mechanical properties, such as modulus and strength, of the final laminate product. While many studies have focused on predicting the mechanical properties of these fiber-reinforced composites, the environment and cycles is usually not considered. In this work, a data-driven method is applied to various uni-directional carbon fiber laminates to investigate the effects of curing environments on mechanical properties such as strength and modulus in longitudinal and transverse directions. Several different curing environments are considered: cold temperature dry (CTD) at -65 °F, room temperature dry (RTD) at 75 °F, elevated temperature dry (ETD) at 200 °F, elevated temperature wet (ETW) at 200 °F, and a second elevated temperature wet (ETW2) environment at 250 °F. To achieve these conditions, specific cycles have been designed to obtain maximum control over the rate in which temperature, pressure and vacuum are both applied to the material. This controlled manufacturing is applied to a variety of resins including Solvay MTM45-1 and Hexcel 8552 are studied, as well as 12K-AS4 and IM7 carbon fibers. We have conducted statistical and exploratory data analyses to identify trends, using data from the National Center for Advanced Materials Performance (NCAMP). Results show that high curing temperatures can yield stronger composites, and that the variability in material property values may be resistant to temperature for some resins but not all. Additionally, using supervised machine learning techniques, we develop and compare decision tree regressive models considering varying curing environments to predict the strength and modulus of these materials in both longitudinal and transverse directions. Specifically, random forest regressive and multi-classification models are used both regressively to predict material properties and for inverse design purposes to design a composite optimally. This work establishes a statistical framework to analyze complex empirical data for both inference and insight for optimal designs. Additionally, it lays groundwork for the future of big data in advanced material design.