【摘要】 破片作为战斗部的主要毁伤元素，在防护工程、武器设计及弹道学研究中具有重要意义。钨球因其高密度、高硬度和优良的机械性能，被广泛用于侵彻Q235钢板的实验研究，以评估防护材料的抗毁伤能力。然而，传统实验方法因成本高、周期长及数据获取有限，难以满足精确预测的需求。随着计算力学和机器学习技术的发展，基于数据驱动的模型成为研究侵彻机理和预测极限弹道速度的有效工具。针对钨球侵彻Q235钢板的复杂非线性关系，提出了一种基于前馈神经网络(FNN)与高斯线性回归(GLR)相结合的混合模型，并采用遗传算法优化网络结构，以提升预测精度。研究通过数值仿真和实验测试获取大量数据，分析破片尺寸、靶板厚度及入射速度等关键因素对侵彻行为的影响，并利用机器学习方法构建高效预测模型。结果表明，该混合模型能够准确预测弹道极限速度及侵彻深度，为防护材料优化设计提供了重要参考。研究不仅提高了侵彻性能评估的精度和可靠性，也为智能化毁伤预测提供了新思路。结果表明，BP-GR混合网络的计算误差约为3.9%,优于理论计算的5.67%。所提出的混合方法与传统理论相比，精度提高了1.77%,突出了其在弹道应用中更准确预测的潜力。更多还原

【Abstract】 As the main damage element of warhead, fragment plays an important role in protection engineering, weapon design and ballistics research. Because of its high density, high hardness and excellent mechanical properties, tungsten ball is widely used in the experimental study of penetrating Q235 steel plate to evaluate the damage resistance of protective materials. However, due to high cost, long cycle and limited data acquisition, traditional experimental methods are difficult to meet the needs of accurate prediction. With the development of computational mechanics and machine learning techniques, data-driven models have become an effective tool to study penetration mechanism and predict the ultimate ballistic velocity. In this paper, a hybrid model based on the combination of feedforward neural network(FNN) and Gaussian linear regression(GLR) is proposed to solve the complex nonlinear relationship of tungsten ball penetration of Q235 steel plate, and genetic algorithm is used to optimize the network structure to improve the prediction accuracy. A large amount of data is obtained through numerical simulation and experimental testing, and the influence of key factors such as fragment size, target plate thickness and incident velocity on the penetration behavior is analyzed, and an efficient prediction model is constructed by using machine learning method. The results show that the hybrid model can accurately predict the ballistic limit velocity and penetration depth, which provides an important reference for the optimal design of protective materials. The research not only improves the accuracy and reliability of penetration performance evaluation, but also provides a new idea for intelligent damage prediction. The results show that the calculation error of BP-GR hybrid network is about 3.9%, which is better than the theoretical calculation of 5.67%. The proposed hybrid method achieves a 1.77% improvement in accuracy compared to the traditional theory, highlighting its potential for more accurate prediction in ballistic applications.更多还原