Application of cascade neural networks for modeling and forecasting the characteristics of energy-yielding materials
Year: 2026
Type: article
Abstract: Modern problems in the field of materials science of energy-yielding materials such as explosives, require developing highly accurate and effective methods for predicting their detonation and sensitivity characteristics. Traditional approaches based on experimental research and theoretical calculations are often resource-hungry and do not give the opportunity to identify complex nonlinear relationships between the structure, composition and properties of materials. Due to this, machine learning methods open up new opportunities for building predictive models that can take into account the multifactorial influence of physico-chemical parameters. However, the task of ensuring not only high accuracy, but also the physical interpretability of the results remains relevant. The aim of the study is to increase the accuracy of predicting detonation and sensitivity characteristics of energy-yielding materials through the development and application of a cascade neural network approach that provides physical interpretability of the results and the possibility of directional design of new materials. Materials and methods. The object of the study is individual and composite explosives. The subject of the study is dependence of detonation characteristics on physico–chemical parameters, detected using cascade neural networks. The cascade method of neural networks is used, in which sequential training of three networks makes it possible to effectively eliminate the noise of experimental data and achieve high prediction accuracy. The study was conducted on a sample of 32 compounds with the analysis of 11 physico-chemical parameters in each compound. Results. A mathematical model based on cascade neural networks is proposed for energy-yielding materials, demonstrating exceptional accuracy in predicting detonation characteristics: detonation velocity (R² = 0,996), detonation pressure (R² = 0,993), impact sensitivity (R² = 0,862) and friction sensitivity (R² = 0.877). The constructed models are established to adequately reflect fundamental physical patterns, including the quadratic dependence of detonation pressure on density and the presence of an optimum oxygen balance in the range of –30...–10%. It is experimentally confirmed that introduction of nanocarbon additives (multilayer carbon nanotubes, single-layer carbon nanotubes) leads to a systemic improvement in detonation characteristics of composite materials. Conclusions. The developed cascade neural network approach demonstrates high efficiency for predicting detonation characteristics of energy-yielding materials and ensures physical interpretability of the results. The models obtained make it possible to correctly reproduce the fundamental laws of explosives physics and can be used for targeted designing new materials with specified properties. The established quantitative relationships between physico-chemical parameters and detonation characteristics open up new opportunities for reducing time and resources in the development of promising energy materials.
Institution: Chuvash State University
Language: English
Subfield: Materials Chemistry
Field: Materials Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7203548218