Total impulse is one of the most important parameters which is highly related to the performance of solid rocket motors. Predicting the total impulse accurately is necessary for both design and operation purposes. However, the traditional methods greatly rely on expert knowledge and are less capable of analyzing modern sophisticated equipment. In this paper, a deep learning-based total impulse prediction method is proposed for the ignition process of solid rocket motor. A CNN-LSTM-Attention deep neural network model is established, which can automatically process raw data for feature extraction, efficiency improvement and prediction with high accuracy. Practical rocket data are used for validations which are collected in the ignition process. We compared the proposed method with the other popular algorithms to verify the effectiveness and superiority of this method. The results show that the proposed data processing and prediction method can achieve promising performance. The best result of average percentage error on the test set is below 2%. The dependency of the deep learning-based method on the data amount is largely reduced by using the downsampling method in data processing. In this way, the proposed method is available even with very few training data, which has good application prospects in engineering problems and provides an approach for combining artificial intelligence and solid rocket motor research.

Solid rocket motors (SRMs) have been popularly used in the current aerospace industry. Performance indicators, such as pressure and thrust, are of great importance for rocket monitoring and design. However, the measurement of such signals requires high economic and time costs. In many practical situations, the thrust measurement error is large and requires manual correction. In order to address this challenging problem, a lightweight RepVGG-based cross-modality data prediction method is proposed for SRMs. An end-to-end data prediction framework is established by transforming data across different modalities. A novel RepVGG deep neural network architecture is built, which is able to automatically learn features from raw data and predict new time-series data of different modalities. The effectiveness of the proposed method is extensively validated with the field SRM data. The accurate prediction of the thrust data can be achieved by exploring the pressure data. After calculation, the percentage error between the predicted data and the actual data is less than 5%. The proposed method offers a promising tool for cross-modality data prediction in real aerospace industries for SRMs.