Cai Zhongman, Ji Hongchao, Pei Weichi, Tang Xuefeng, Xin Long, Lu Yonghao, Li Wangda
College of Mechanical Engineering, North China University of Science and Technology, 21 Bohai Road, Caofeidian Xincheng, Tangshan 063210, China.
National Center for Materials Service Safety, University of Science and Technology Beijing, Beijing 100083, China.
Materials (Basel). 2020 Mar 12;13(6):1282. doi: 10.3390/ma13061282.
Based on an 33Cr23Ni8Mn3N thermal simulation experiment, the application of an artificial neural network (ANN) in thermomechanical processing was studied. Based on the experimental data, a microstructure evolution model and constitutive equation of 33Cr23Ni8Mn3N heat-resistant steel were established. Stress, dynamic recrystallization (DRX) fraction, and DRX grain size were predicted. These models were evaluated by a variety of statistical indicators to determine that these models would work well if applied in predicting microstructure evolution and that they have high precision. Then, based on the weight of the ANN model, the sensitivity of the input parameters was analyzed to achieve an optimized ANN model. Based on the most widely used sensitivity analysis (SA) method (the Garson method), the input parameters were analyzed. The results show that the most important factor for the microstructure of 33Cr23Ni8Mn3N is the strain rate ( ε ˙ ). For the control of the microstructure, the control of the ε ˙ is preferred. ANN was applied to the development of processing map. The feasibility of the ANN processing map on austenitic heat-resistant steel was verified by experiments. The results show that the ANN processing map is basically consistent with processing map based on experimental data. The trained ANN model was implanted into finite element simulation software and tested. The test results show that the ANN model can accurately expand the data volume to achieve high precision simulation results.
基于33Cr23Ni8Mn3N热模拟实验，研究了人工神经网络（ANN）在热机械加工中的应用。基于实验数据，建立了33Cr23Ni8Mn3N耐热钢的微观组织演变模型和本构方程。预测了应力、动态再结晶（DRX）分数和DRX晶粒尺寸。通过多种统计指标对这些模型进行评估，以确定这些模型在预测微观组织演变时是否有效且具有高精度。然后，基于人工神经网络模型的权重，分析输入参数的敏感性以获得优化的人工神经网络模型。基于最常用的敏感性分析（SA）方法（加森方法），对输入参数进行了分析。结果表明，影响33Cr23Ni8Mn3N微观组织的最重要因素是应变速率（ε˙）。对于微观组织的控制，优先控制ε˙。将人工神经网络应用于加工图的开发。通过实验验证了人工神经网络加工图在奥氏体耐热钢上的可行性。结果表明，人工神经网络加工图与基于实验数据的加工图基本一致。将训练好的人工神经网络模型植入有限元模拟软件并进行测试。测试结果表明，人工神经网络模型可以准确扩展数据量以获得高精度的模拟结果。
