基于Adaboost的填充式防护结构超高速撞击损伤预测

填充式防护结构的显式弹道极限方程在对弹丸进行超高速撞击损伤预测时，由于填充材料、填充方式的不同，会导致预测结果与实测数据存在一定偏差。对此，采用机器学习方式将该问题转化为二分类问题，以碰撞过程中的弹丸撞击参数、防护结构参数作为分类特征，构建了基于Adaboost的填充式防护结构超高速撞击损伤预测模型。该模型以分类回归树（CART）作为弱分类器，通过对一系列弱分类器的加权组合生成强分类器，并通过对训练样本的循环使用，实现了小样本集下的撞击损伤预测。实验结果表明，建立的Adaboost预测模型对填充式防护结构的超高速撞击损伤具有良好的预测效果，总体预测率与安全预测率相比于NASA的弹道极限方程均提高了14.3%，具有更强的通用性。通过不同训练样本规模下的交叉检验，证明了该模型具有良好的鲁棒性与准确性。 

Hypervelocity impact damage prediction of stuffed Whipple shield based on Adaboost

The explicit ballistic limit equation of stuffed Whipple shield may cause some deviations between the prediction results and the measured data when the projectile is subjected to hypervelocity impact damage prediction because of different filling materials and filling methods. In this regard, the machine learning method is used to transform the problem into a binary problem. The projectile impact parameters and protective structure parameters in the collision process are used as the classification features to construct a hypervelocity impact damage prediction model of stuffed Whipple shield based on Adaboost. The model uses the classification and regression tree (CART) as a weak classifier to generate a strong classifier by weighted combination of a series of weak classifiers. Through the cyclic use of training samples, the impact damage prediction under a small sample set is achieved. The experimental results show that the established Adaboost prediction model has good prediction effect on the hypervelocity impact damage of stuffed Whipple shield. Both the totality prediction rate and the safety prediction rate of Adaboost prediction model increase by 14.3% compared with NASA's ballistic limit equation, and the established model has more versatility. Cross check under different training sample sizes proves that the model has good robustness and accuracy.