Machine learning for burning rate prediction of solid propellants

Abstract

The burn rate is a key performance parameter in the development of new propellant formulations. However, its experimental determination is highly time- and resource-intensive, which demands faster, more economical predictive methods prior to experimentation. In this study, a machine learning (ML) approach for predicting the burning rate of composite propellants is presented. The dataset comprises 225 unique propellant formulations derived from solid rocket motor processing. A set of input features, including compositional parameters, chemical properties, and processing conditions, was selected for model development. Various linear and non-linear models were trained, among which support vector regression with a linear kernel achieved the best performance, yielding a root mean squared error of 1.12±0.20 relative to experimental burning rates. Other algorithms, such as neural networks and Gaussian processes, showed comparable predictive capability. Feature analysis revealed contributions consistent with established domain knowledge. The practical utility of this approach was further demonstrated by screening a large number of virtual formulations with the trained model; those with predicted burning rates close to a target value were experimentally validated. The proposed workflow offers a promising pathway to accelerate the discovery of new propellant compositions while enhancing cost-effectiveness, efficiency, reliability, and intelligence in the development process.