Domain Adaptive Based Thrust Prediction Method for Solid Rocket Motor

The prediction of solid rocket motor performance is crucial for both rocket monitoring and engineering design programs. Accurate thrust prediction significantly enhances rocket reliability and reduces development costs in terms of time and economics. Traditionally, the limited data obtained from solid rocket motor tests poses a challenge in achieving the required prediction accuracy using neural networks. In this paper, we propose a deep learning-based domain adaptive method to address the issue of insufficient sample data in solid rocket motor testing. Our method leverages the Maximum Mean Discrepancy (MMD) technique in transfer learning to tackle domain shifting problems. It approximates the distribution of high and low-temperature domains within the feature representation (hidden layer information of the network) learned by the neural network. The results demonstrate the effectiveness of our proposed method, showing a 6.125% improvement in prediction accuracy over traditional neural networks. This validates the efficacy and superiority of our approach. Hence, our method offers valuable insights into handling small sample data in the field of solid rockets and holds promising applications.