The most accurate ANN learning algorithm for FEM prediction of mechanical performance of alloy A356
The most accurate ANN learning algorithm for FEM prediction of mechanical performance of alloy A356
复制标题
DOI：
10.4149/km_2012_1_25
复制
发表时间：
2012-01-01
影响因子：
0.7
通讯作者：
Razavi, M.
中科院分区：
文献类型：
--
作者：
Shabani, M. O.;Mazahery, A.;Razavi, M.
为了找到最准确的预测屈服应力，UTS和延伸率，不同的训练算法的学习性能的神经网络的影响进行了研究。在神经网络的训练和测试模块中，以不同的一次和二次枝晶间距作为输入，以屈服应力、极限拉伸强度和延伸率作为输出。准备好训练集后，使用不同的训练算法、隐藏层和隐藏层中的神经元数量来训练神经网络。测试集用于在学习结束时检查每个训练算法的系统准确性。结果表明，Levenberg-Marquardt学习算法对A356合金的屈服应力、极限拉伸强度和延伸率的预测效果最好。
In order to discover the most accurate prediction of yield stress, UTS and elongation percentage, the effects of various training algorithms on learning performance of the neural networks were investigated. Different primary and secondary dendrite arm spacings were used as inputs, and yield stress, UTS and elongation percentage were used as outputs in the training and test modules of the neural network. After the preparation of the training set, the neural network was trained using different training algorithms, hidden layers and neuron numbers in hidden layers. The test set was used to check the system accuracy of each training algorithm at the end of learning. The results show that Levenberg-Marquardt learning algorithm gave the best prediction for yield stress, UTS and elongation percentage of A356 alloy.