A Novel Method for Damage Prediction of Stuffed Protective Structure under Hypervelocity Impact by Stacking Multiple Models

Abstract

Adopting appropriate protective structures to resist the impact of micrometeoroid or space debris on spacecraft is a valuable research content of space engineering. After the single-layer board and Whipple protective structure, the stuffed protective structure has become one of the research focuses in recent years. However, due to the different filling materials and filling modes, the prediction of hypervelocity impact damage of projectile by using the explicit ballistic limit equation of the stuffed protective structure will lead to some deviation between the predicted results and the measured data. In this paper, the problem is transformed into a binary classification problem, which is characterized by projectile impact parameters and protective structure parameters. And a novel method is proposed to predict the hypervelocity impact damage of the stuffed protective structure under the stacking ensemble learning framework. In this method, the hypervelocity impact damage data set of the stuffed protective structure is firstly divided into specific data subsets. Extreme Gradient Boosting (XGBoost), Random Forest (RF), Support Vector Machine (SVM), and K-Nearest Neighbor (KNN) are taken as the base models of the stacking ensemble learning framework. And each base model is trained by using the divided sub-datasets. Thus the meta-features with stronger generalization ability are extracted. Then, the Long Short Term Memory (LSTM) network is used as the meta-model of the stacking ensemble learning framework, and the meta-features extracted before are used to train it. In this way, the high-precision prediction of the impact damage of the stuffed protective structure is realized under the small data set. The experimental results show that the stacking ensemble learning model has a good prediction effect on the impact damage of the stuffed protective structure. The cross-validation under different training sample sizes further proves that the stacking ensemble learning model has excellent robustness and accuracy.