The ductile fracture behavior of metallic materials is usually coupled with complex stress states. In order to describe the facture behavior of metallic materials under a broad range of stress states, a modified Gurson-Tvergaard-Needleman damage model was proposed by Wei Jiang and has been used to simulate the ductile fracture of 2024-T3 aluminum alloy. However, it was a heavy and trifle job to determine the unknown parameters of the model because of the large number of them and inefficient identification strategy. In this research, we propose an efficient parameters identification strategy of this new damage model based on machine learning algorithm. This strategy combines resilient back-propagation neuro network with genetic algorithm and simulations were implemented in terms of ABAQUS/Explicit. The same strategy of optimization was also used to reduce simulation time. Damage model parameters of 2024-T3 aluminum alloy were identified from the experimental force-displacement curves of the specimens exhibiting high stress triaxiality and negative triaxiality. The identified parameters were verified by using the test results of the specimens providing low and zero stress triaxialities.

金属材料的韧性断裂行为通常伴随着复杂的应力状态。为了描述金属材料在大范围应力状态下的断裂行为，Wei Jiang提出了一种改进的Gurson-Tvergaard-Needleman损伤模型，并用于模拟2024-T3铝合金的延性断裂。然而，由于模型数量众多且识别策略效率低下，确定模型的未知参数是一项繁重而繁琐的工作。在这项研究中，我们提出了一种基于机器学习算法的新损伤模型的有效参数识别策略。该策略将弹性反向传播神经网络与遗传算法相结合，并根据ABAQUS / Explicit进行了仿真。相同的优化策略也用于减少仿真时间。从高应力三轴性和负三轴性的试样的实验力-位移曲线确定了2024-T3铝合金的损伤模型参数。通过使用提供低应力和零应力三轴性的试样的测试结果来验证所确定的参数。