Rapid Detection of Stabilizer Content in Double-Base Propellant Based on Artificial Neural Network Combined With Near-Infrared Spectroscopy

Abstract

During long-term storage, double-base propellants are prone to chemical decomposition of internal nitrate esters, leading to decreased burn rate, reduced strength, and degraded ballistic performance. Adding an appropriate amount of Centralite-II is crucial for ensuring storage safety. This study proposes a novel method combining near-infrared spectroscopy (NIRS) with artificial intelligence to rapidly and non-destructively detect the content of Centralite-II in double-base propellants. The optimal modeling wavelength ranges of 4000–4600 cm⁻¹ and 5700–6100 cm⁻¹ were identified, and the raw spectral data were preprocessed using standard normal variate (SNV) transformation to improve the signal-to-noise ratio. Principal component analysis (PCA) was then applied to reduce data dimensionality, and the first three principal components were used as inputs for a backpropagation (BP-ANN) neural network. The resulting PCA-BP-ANN model showed excellent performance on the training set, with an $R_c^2$ of 0.9830 and an $\text{RMSEC}$ of 0.0376%. During independent validation, the model demonstrated strong generalization ability, achieving an $R_p^2$ of 0.9824 and an $\text{RMSEP}$ of 0.3179%, comparative analysis with other models, including BP, PLS, ELM, SVR, and LSTM, indicated that the PCA-BP-ANN model exhibited superior prediction accuracy and generalization capability. This method provides a rapid and non-destructive approach for assessing the stabilizer content in double-base propellants and expands the application of NIRS and AI techniques in the field of energetic materials.