(1. 燕山大学 先进锻压成形技术与科学教育部重点实验室，秦皇岛 066004；
2. 燕山大学 河北省特种运载装备重点实验室，秦皇岛 066004)
摘 要: 为有效预测AA7075-T6板材变形破裂问题，设计10种不同应力状态的板材拉伸试样，通过方程组法获取BP神经网络的样本数据，建立基于神经网络与遗传算法(BP+GA法)的韧性断裂准则参数预测模型，并依据方程组法最优试样组合方案以及优化后的断裂参数，绘制AA7075-T6板材成形极限曲线。通过缺口试样误差评估比较方程组法和BP+GA法的断裂预测精度，并应用半球形刚模胀形试验对方程组法和BP+GA法两种断裂参数标定方法绘制的成形极限曲线(FLC)进行验证。结果表明：方程组法筛选后的最佳试样组合方案接近于BP+GA法搜索得到的全局最优解；通过BP+GA法绘制的AA7075-T6板材理论成形极限曲线为成形极限实测数据点集的下轮廓，预测结果趋近安全；而缺少平面应变至双向等拉区域的试验样本导致理论FLC产生较大差距，从而反映了Lou-Huh准则参数求解对测试试样应力状态具有较高的敏感性。研究结果为高强铝板断裂理论参数分析和成形极限预测提供了借鉴和数据依据。
关键字: 韧性断裂准则；BP神经网络；遗传算法；高强铝板；成形极限预测
（1. Key Laboratory of Advanced Forging and Stamping Technology and Science, Ministry of Education, Yanshan University, Qinhuangdao 066004, China;
2. Hebei Key Laboratory of Special Delivery Equipment, Yanshan University, Qinhuangdao 066004, China）
Abstract:In order to effectively predict AA7075-T6 sheet deformation problem, the 10 sheet tensile specimens for different stress states were designed, according to the equations method to obtain sample data of BP neural network. A fracture parameter prediction model of ductile fracture criterion based on neural network and genetic algorithm (BP+GA) was established. The forming limit curves of AA7075-T6 sheet were drawn based on the optimal specimen option by the equations method and the fracture parameters optimized by the BP+GA method. The fracture prediction accuracies of equations method and BP+GA method were compared by evaluating the fracture prediction error of notched specimen, and the forming limit curves drawn by the equations method and the BP+GA method were verified by punch-stretch test. The results show that the optimal specimen option selected by the equations method is close to the global optimal solution obtained by the BP+GA method. The theoretical forming limit curve (FLC) of AA7075-T6 sheet drawn by BP+GA method is the lower profile of the experimental data point set, and the predicted result is safe. However, the lack of tensile specimens from plane strain to biaxial-equal tension stress regions results in a large gap between the theoretical FLC and test data, which reflects that the parameter calculation for Lou-Huh criterion has a high sensitivity to the stress state of the test specimen. The research result provides reference and data basis for fracture parameters analysis and forming limit prediction of high strength aluminum sheet.
Key words: ductile fracture criterion; BP neural network; genetic algorithm; high strength aluminum sheet; forming limit prediction
