Lightweight RepVGG-Based Cross-Modality Data Prediction Method for Solid Rocket Motors.

Area of Science:

• Aerospace Engineering
• Data Science
• Machine Learning

Background:

• Solid rocket motors (SRMs) are crucial in aerospace, but performance monitoring (pressure, thrust) is costly and error-prone.
• Accurate thrust measurement in SRMs often requires expensive equipment and manual correction, impacting efficiency.
• Existing methods face challenges in real-time, cost-effective performance monitoring for SRMs.

Purpose of the Study:

• To develop a novel, lightweight, cross-modality data prediction method for SRMs.
• To establish an end-to-end framework for predicting SRM performance indicators using deep learning.
• To reduce the economic and time costs associated with SRM thrust measurement.

Main Methods:

• A RepVGG-based deep neural network architecture was designed for feature learning from raw data.
• The method transforms data across different modalities for predictive modeling.
• An end-to-end framework was implemented for time-series data prediction.

Main Results:

• The proposed method accurately predicts SRM thrust data using pressure data.
• The prediction model achieved a percentage error of less than 5% compared to actual data.
• Validation was performed using field SRM data, demonstrating practical effectiveness.

Conclusions:

• The RepVGG-based cross-modality prediction offers a promising solution for SRM performance monitoring.
• This approach can significantly improve the accuracy and efficiency of thrust data prediction.
• The method provides a valuable tool for real-world aerospace applications involving SRMs.