Investigation on enhancement of filler dispersion and prediction of mechanical behavior of hexagonal boron nitride/epoxy nanocomposites through machine learning and deep learning models

Abstract Two‐dimensional hexagonal boron nitride (hBN) based nanocomposites exhibit excellent mechanical and thermal properties for various electronics, automotive, and aerospace applications. The present work gives a novel approach to fabricating hexagonal boron nitride/epoxy nanocomposites using isopropanol and dimethyl ketone as dispersants in two different routes and to predict mechanical characteristics employing deep learning and machine learning models. Nanocomposites were fabricated by employing casting techniques with varying concentrations of hBN, spanning from 0.25 to 1 wt%, utilizing dispersing solvents. The nanocomposites were analyzed for mechanical behavior, highlighting a notable improvement in the mechanical properties at 0.5 wt% isopropanol dispersed hBN. It showcased 61% improvement in tensile strength, 38.41% increase in flexural strength, and 35.80% increase in flexural modulus respectively as compared to the pristine epoxy. The Halpin Tsai analytical model showed agreement with the elastic modulus calculated experimentally. The fractured SEM micrograph supported the improved dispersion of the hBN nanocomposite. Thermal stability of 0.5 wt% isopropanol dispersed hBN/epoxy nanocomposite revealed an improvement by 8°C at 50% degradation as compared to the pristine epoxy. Linear regression, random forest regression, support vector regression, and deep neural network (DNN) were employed to predict values. DNN proved better results by showcasing low prediction loss and high R 2 values (0.99468–0.99966). Highlights hBN/epoxy nanocomposite with isopropanol and dimethyl ketone as dispersants. 0.5% hBN loading in isopropanol exhibited improved mechanical characteristics. The Halpin Tsai model was employed for evaluating theoretical elastic modulus. Improved thermal stability and filler dispersion by optimized hBN/epoxy combination. Deep neural network showed higher R 2 value and lower prediction loss.