基于人工神经网络的超级马氏体不锈钢淬火力学性能预测
Prediction About Mechanical Properties of Quenched Supermartensitic Stainless Steel by ANN
刘 环, 邹德宁, 闫东娜, 于军辉
点击：5785次 下载：0次
作者单位：西安建筑科技大学 冶金工程学院, 陕西 西安 710055
中文关键字：超级马氏体不锈钢； 淬火工艺； 力学性能； 人工神经网络
英文关键字：SMSS; quenching technique; mechanical property; ANN
中文摘要：利用人工神经网络(ANN)方法建立了超级马氏体不锈钢(SMSS)淬火工艺参数与力学性能的预测模型。模型输入单元为淬火温度、保温时间和冷却方式，输出单元为抗拉强度、屈服强度和伸长率；网络为3-9-3结构，动量因子为0.2，采用提前终止法与Levenberg-Marquardt算法相结合训练网络，以实验结果验证网络的可靠性。预测结果表明，抗拉强度、屈服强度和伸长率相对误差绝对值的最大值分别为2.2050%、1.4393%和8.4211%。该模型可为SMSS热处理工艺制定提供参考依据。
英文摘要：The mechanical properties of quenched supermartensitic stainless steel (SMSS) were predicted using artificial neural network (ANN) model. Quenching temperature, aging time and type of cooling were employed as inputs while yield stress, tensile strength and elongation were taken as outputs. The method to combine early stopping algorithm with Levenberg-Marquardt algorithm was employed to train ANN model. Then the experiment results were used to check the model accuracy. The optimal network architecture is considered to be 3-9-3 with momentum factor 0.2. The results show that the biggest absolute relative errors (ARE) of yield stress, tensile strength and elongation are 2.2050%, 1.4393% and 8.4211%, respectively. The model can be used as a reference for determining SMSS heat treatment technological parameters.