Scholarlink

<p>Accurate and efficient prediction of concrete anti-penetration behavior enables rapid safety assessments of protective structures. Concrete penetration involves complex multi-physical coupling (material dynamics, stress wave propagation, strain rate effects), while scarce test data and noise interference limit empirical formulas and phenomenological models. Numerical simulations also suffer from low efficiency. Thus, developing a model that can rapidly and precisely forecast the depth of concrete penetration is crucial. This paper proposes Back Propagation Neural Network (BP) combining Genetic Algorithm (GA) and Adaptive Boosting Algorithm (AdaBoost). First, a bias-based outlier detection method is used to screen the outliers in the sample data. The GA is used to optimize the weights and thresholds of the BP. The optimized GA-BP serves as a weak learner. Multiple weak learners are then combined using the AdaBoost to create a strong learner model. Secondly, the prediction effects of BP model, GA-BP model and GA-BP-AdaBoost model are compared on the same dataset, the results show that the coefficients of determination of the three are 0.79, 0.81, and 0.9, respectively, which proves that the GA-BP-AdaBoost model significantly improves the prediction accuracy and has obvious advantages compared with traditional methods. Finally, penetration tests were performed using a 20&#xa0;mm ballistic gun with sub-caliber techniques, firing tungsten alloy projectiles at speeds from 168 to 266&#xa0;m/s. The test data were input into the model for verification. The results showed that the predicted values were close to the experimental values, which indicated that the model has good practical value.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Machine Learning-Driven High-Accuracy Prediction of Concrete Penetration Depth

  • 期刊论文

  • 发表日期：

    2026年2月21日

查看全文
  • P. Li,
  • J. Zhong

摘要

Accurate and efficient prediction of concrete anti-penetration behavior enables rapid safety assessments of protective structures. Concrete penetration involves complex multi-physical coupling (material dynamics, stress wave propagation, strain rate effects), while scarce test data and noise interference limit empirical formulas and phenomenological models. Numerical simulations also suffer from low efficiency. Thus, developing a model that can rapidly and precisely forecast the depth of concrete penetration is crucial. This paper proposes Back Propagation Neural Network (BP) combining Genetic Algorithm (GA) and Adaptive Boosting Algorithm (AdaBoost). First, a bias-based outlier detection method is used to screen the outliers in the sample data. The GA is used to optimize the weights and thresholds of the BP. The optimized GA-BP serves as a weak learner. Multiple weak learners are then combined using the AdaBoost to create a strong learner model. Secondly, the prediction effects of BP model, GA-BP model and GA-BP-AdaBoost model are compared on the same dataset, the results show that the coefficients of determination of the three are 0.79, 0.81, and 0.9, respectively, which proves that the GA-BP-AdaBoost model significantly improves the prediction accuracy and has obvious advantages compared with traditional methods. Finally, penetration tests were performed using a 20 mm ballistic gun with sub-caliber techniques, firing tungsten alloy projectiles at speeds from 168 to 266 m/s. The test data were input into the model for verification. The results showed that the predicted values were close to the experimental values, which indicated that the model has good practical value.