Accelerated Discovery of Graphene Kirigami with an Enhanced Elastocaloric Effect via Machine Learning

Recent studies have examined the elastocaloric response of graphene kirigami (GK) and shown how it may be tailored through geometric design. This tunability makes GK a promising platform for applications in nanoscale solid-state thermal devices. In this work, we combine molecular dynamics (MD) simulations and machine learning (ML) to explore how GK geometries affect the elastocaloric coefficient (ECC), defined as the adiabatic ratio between temperature change and applied tensile stress. A data set of 16,807 GK configurations was generated through systematic cut patterns and evaluated via MD at room temperature. Using this data, both classical and deep-learning models were trained, with a convolutional neural network (CNN) achieving the best performance (RMSE = 0.064 K GPa^-1; <i>R</i>² = 0.96). Model-guided optimization identified high-ECC designs 10 times faster than random search, demonstrating the power of ML-assisted strategies for the accelerated discovery of advanced elastocaloric materials.