EMNN: Data-driven Kinetic Modelling of Energetic Materials from Differential Scanning Calorimetry Experiments

A comprehensive understanding of the thermal decomposition kinetics of energetic materials (EMs) is essential for enhancing their application in aerospace systems. Previous studies have employed chemical reaction neural networks (CRNNs) to infer decomposition kinetics from thermogravimetric (TG) data; however, these models fail to capture heat release profiles, which are critical for predicting thermal runaway behavior. Here, we propose an energetic material neural network (EMNN) that integrates physical constraints with data-driven optimization to derive multistep reaction kinetics from differential scanning calorimetry (DSC) measurements. After validation using a benchmark elementary reaction, the EMNN framework is applied to develop decomposition models for 2,4,6,8,10,12-hexanitro-2,4,6,8,10,12-hexaazaisowurtzitane (CL-20) and dihydroxylammonium 5,5′-bistetrazole-1,1′-diolate (TKX-50). Both models demonstrate superior predictive accuracy for DSC experiments compared to CRNN-based approaches. The successful implementation of the EMNN underscores its potential to improve thermal runaway modeling for EMs and other complex kinetic systems.