Open Access

ARTICLE

Rheological Properties of Solid Rocket Propellants Based on Machine Learning

Minghai Zheng¹, Zhaoxia Cui^1,*, Jiang Liu¹, Jianjun Li²

1 School of Mechanical Engineering, Inner Mongolia University of Technology, Hohhot, 010051, China
2 Inner Mongolia Aerospace Hongxia Chemical Co., Ltd., Hohhot, 010076, China

* Corresponding Author: Zhaoxia Cui. Email:

(This article belongs to the Special Issue: Machine Learning-Assisted Structural Integrity Assessment and Design Optimization under Uncertainty)

Computer Modeling in Engineering & Sciences 2025, 145(1), 431-455. https://doi.org/10.32604/cmes.2025.071913

Received 15 August 2025; Accepted 29 September 2025; Issue published 30 October 2025

Abstract

To accurately depict the strong nonlinear relationship between the viscosity of propellant slurry and shear rate, premix time, and temperature, and to improve the prediction accuracy, based on the sample preparation and experimental measurement of a certain type of propellant, viscosity data under multiple working conditions were obtained as the basic data for the research. By comparing typical models such as support vector regression and random forest, it was found that although the traditional BP neural network was superior to the both, its accuracy was still insufficient. Based on this, a BP model co-optimized by the Sparrow Search Algorithm (SSA) and the Genetic Algorithm (GA) is proposed. The global search of SSA and the local convergence of GA are utilized to conduct dual optimization of the initial weights and thresholds of the BP network, and the training is completed based on the measured shear rate, temperature and time data. Further, the Box-Behnken response surface design is adopted to transform the output of the neural network into a quadratic explicit function of viscosity and multiple factors. The results show that the SSA-GA-BP model achieves a determination coefficient of R² = 0.948, compared with R² = 0.628 for the traditional BP neural network, while the root mean square error (RMSE) is reduced from 1093.99 to 154.75. Within the key pouring viscosity range of 200–600 Pa·s, the prediction deviation remains within ±5%, and the overall prediction variance is improved by more than 100%. The polynomial quantification obtained from the response surface reveals the dominant role of shear rate and its interaction with temperature and time. The fitting curve is more in line with the experimental trend than the traditional constitutive model. The constructed explicit function can be directly embedded in Computational Fluid Dynamics (CFD) for pouring process simulation and has good engineering application value.

---

Keywords

---