An improved GAN-based data augmentation model for addressing data scarcity in SRMs

In the domain of solid rocket motors (SRMs), acquiring experimental data is a costly endeavor, and fully simulating flight conditions presents significant challenges, making it difficult to gather a sufficient amount of data for each operational scenario. The application of deep learning architectures in the field of SRMs is further complicated by issues of data scarcity and dataset imbalance. To address these challenges, this paper proposes an enhanced generative adversarial network (GAN)-based data augmentation model, referred to as CWGAN-GP-T. This model aims to effectively augment and balance the experimental dataset of SRMs, thereby improving the regression prediction performance of deep learning architectures utilizing SRM data. By leveraging the CWGAN-GP-T framework, pseudo-samples are generated based on the experimental dataset from the SRMs ignition process. The results demonstrate the efficacy of the proposed method in expanding the dataset and ensuring a more balanced distribution of samples across various operating conditions. Comparative experiments are conducted with alternative regression prediction methods and GAN-based data augmentation techniques, validating the superior accuracy of deep convolutional neural networks in predicting the performance of SRMs using datasets augmented by the CWGAN-GP-T method. These findings highlight the exceptional generative capabilities of the proposed model and emphasize its potential impact on advancing research in the field of SRMs.