基于人工神经网络和M_s点的钢的成分反向设计
|
摘要
本文基于人工神经网络(ANN)和马氏体转变开始温度(M_s点)对钢的成分进行反向设计。首先设计了径向基函数(RBF)人工神经网络模型,用"舍一法"训练了模型,使其具有良好的预测性能。然后,用训练后的模型对钢的成分进行了反向设计,得到的散点大致分布于45°角平分线附近,统计学指标为:均方差均<0.12,相对均方差均<0.18,拟合值均>1.96,表明基于人工神经网络和M_s点的钢的成分反向设计精度高。最后,用神经网络模型研究了钢中其余元素和M_s点对碳含量预测的影响。计算结果显示:钢中其余元素和M_s点对碳含量预测的影响是非线性的,这主要是钢中各元素、Ms点间存在相互作用造成的。
Based on artificial neural network and martensite transformation starting temperature, this article was to realize the reverse design of the chemical composition of steels. Firstly, radial basis function artificial neural network was established, using the method of "leave-one-out" to practice the model to achieve good prediction performance. Then, using the practiced model to reverse design of the chemical composition of steels, the predicted and measured values distributed along the 45°diagonal in the scatter-plot diagram and their statistical indicators were: all of the MSE < 0.12, all of the MSRE < 0.18, and all of the VOF > 1.96. All results above shows that, the reverse design with a high prediction precision. Finally, the influence of other elements and Mspoints on the carbon content of steels was investigated using the neural network model. The calculation exhibits that the influence of other elements and Mspoints on the prediction of the carbon content of the steels is nonlinear and shows different effects in different ranges. These findings are mainly caused by the interaction between the elements in the steels and the Mspoints.
Based on artificial neural network and martensite transformation starting temperature, this article was to realize the reverse design of the chemical composition of steels. Firstly, radial basis function artificial neural network was established, using the method of "leave-one-out" to practice the model to achieve good prediction performance. Then, using the practiced model to reverse design of the chemical composition of steels, the predicted and measured values distributed along the 45°diagonal in the scatter-plot diagram and their statistical indicators were: all of the MSE < 0.12, all of the MSRE < 0.18, and all of the VOF > 1.96. All results above shows that, the reverse design with a high prediction precision. Finally, the influence of other elements and Mspoints on the carbon content of steels was investigated using the neural network model. The calculation exhibits that the influence of other elements and Mspoints on the prediction of the carbon content of the steels is nonlinear and shows different effects in different ranges. These findings are mainly caused by the interaction between the elements in the steels and the Mspoints.
引文
[1]徐祖耀.马氏体相变与马氏体[M].北京:科学出版社,1999:1-39.
[2]由伟,方鸿生,白秉哲.用反向传播人工神经网络预测低碳低合金钢的马氏体转变开始温度[J].金属学报,2003,39(6):630-634.
[3]由伟.人工神经网络预测新型空冷贝氏体钢的CCT图[D].北京:清华大学材料系,2004.
[4]Senuma T,Suehiro M,Yada H.Mathematical Models for Predicting Microstructural Evolution and Mechanical Properties of Hot Strips[J].ISIJ international,1992,32(3):423-432.
[5]康大韬,郭成熊.工程用钢的组织转变与性能图册[M].北京:机械工业出版社,1992.
[6]汤爱涛,胡红军,杨明波.计算机在材料工程中的应用[M].重庆:重庆大学出版社,2008.
[7]Hecht Nielesn.Neurocomputing[M].Massachussetes:Addison Wesley Publishing Co,Inc,1991:59-63.
[8]赵文雅.基于RBF神经网络的马氏体开始转变温度预测研究[J].热加工工艺,2014,43(6):47-49.
[9]由伟,赵玮玮,赖惠先,等.用人工神经网络研究钢的硬度的影响因素[J].钢铁研究学报,2013,25(1):34-38.
[10]王松林,吴莹,张星,等.基于人工神经网络和硬度的钢的成分反向设计[J].热加工工艺,2014,43(14):67-71.
[11]由伟,白秉哲,方鸿生.钢的连续冷却转变图的神经网络计算模型及预测软件设计[J].金属热处理,2004,29(7):17-20.
[12]张世中.钢的过冷奥氏体转变曲线集[M].北京:冶金工业出版社,1993:22.
[13]The Research Laboratory,United States Steel Company.Atlas of Isothermal Transformation Diagrams[M].Pittsburgh,Kearny,NJ:United States Company Press,1951:3.
[14]The No.1 Iron Factory,Benxi Iron,Steel Corporation,Tsinghua University.The Atlas of Super-Cooling Austenite Transformation Diagrams[M].Benxi:The No.1 Iron and Factory Press,Benxi Iron and Steel Corporation,1978:1.
[15]American Society for Metals.Atlas of Isothermal Transformation and Cooling Transformation Diagrams[M].Metals Park,Ohio:American Society of Metals Press,1977.
[16]张立明.人工神经网络模型及其运用[M].上海:复旦大学出版社,1993.
[17]Eberhart R C,Donbbins R W.Neural Network PC Tools:A Practical Guide[M].New York Academic Press,1990:3.
[18]Narayan V,Abad R,Lopez B,et al.Estimation of Hot Torsion Stress Strain Curves in Iron Alloys Using a Neural Network Analysis[J].ISIJ international,1999,39(10):999-1005.
[19]吴耿峰,王炜,张博锋,等.径向基函数(RBF)神经网络及其应用[J].地震学报,2005,25(2):19-25.
[20]Sung Sau So,Martin Karplus.Evolutionary Optimization in Quantitative Etructure-Activity Relationship:An Application of Genetic Neural Networks[J].Journal of Medicinal Chemistry,1996,39(7):1521-1530.
[21]由伟,白秉哲,方鸿生.用反向传播人工神经网络预测低碳低合金钢的马氏体转变开始温度[J].金属学报,2003,39(6):630-634.
[2]由伟,方鸿生,白秉哲.用反向传播人工神经网络预测低碳低合金钢的马氏体转变开始温度[J].金属学报,2003,39(6):630-634.
[3]由伟.人工神经网络预测新型空冷贝氏体钢的CCT图[D].北京:清华大学材料系,2004.
[4]Senuma T,Suehiro M,Yada H.Mathematical Models for Predicting Microstructural Evolution and Mechanical Properties of Hot Strips[J].ISIJ international,1992,32(3):423-432.
[5]康大韬,郭成熊.工程用钢的组织转变与性能图册[M].北京:机械工业出版社,1992.
[6]汤爱涛,胡红军,杨明波.计算机在材料工程中的应用[M].重庆:重庆大学出版社,2008.
[7]Hecht Nielesn.Neurocomputing[M].Massachussetes:Addison Wesley Publishing Co,Inc,1991:59-63.
[8]赵文雅.基于RBF神经网络的马氏体开始转变温度预测研究[J].热加工工艺,2014,43(6):47-49.
[9]由伟,赵玮玮,赖惠先,等.用人工神经网络研究钢的硬度的影响因素[J].钢铁研究学报,2013,25(1):34-38.
[10]王松林,吴莹,张星,等.基于人工神经网络和硬度的钢的成分反向设计[J].热加工工艺,2014,43(14):67-71.
[11]由伟,白秉哲,方鸿生.钢的连续冷却转变图的神经网络计算模型及预测软件设计[J].金属热处理,2004,29(7):17-20.
[12]张世中.钢的过冷奥氏体转变曲线集[M].北京:冶金工业出版社,1993:22.
[13]The Research Laboratory,United States Steel Company.Atlas of Isothermal Transformation Diagrams[M].Pittsburgh,Kearny,NJ:United States Company Press,1951:3.
[14]The No.1 Iron Factory,Benxi Iron,Steel Corporation,Tsinghua University.The Atlas of Super-Cooling Austenite Transformation Diagrams[M].Benxi:The No.1 Iron and Factory Press,Benxi Iron and Steel Corporation,1978:1.
[15]American Society for Metals.Atlas of Isothermal Transformation and Cooling Transformation Diagrams[M].Metals Park,Ohio:American Society of Metals Press,1977.
[16]张立明.人工神经网络模型及其运用[M].上海:复旦大学出版社,1993.
[17]Eberhart R C,Donbbins R W.Neural Network PC Tools:A Practical Guide[M].New York Academic Press,1990:3.
[18]Narayan V,Abad R,Lopez B,et al.Estimation of Hot Torsion Stress Strain Curves in Iron Alloys Using a Neural Network Analysis[J].ISIJ international,1999,39(10):999-1005.
[19]吴耿峰,王炜,张博锋,等.径向基函数(RBF)神经网络及其应用[J].地震学报,2005,25(2):19-25.
[20]Sung Sau So,Martin Karplus.Evolutionary Optimization in Quantitative Etructure-Activity Relationship:An Application of Genetic Neural Networks[J].Journal of Medicinal Chemistry,1996,39(7):1521-1530.
[21]由伟,白秉哲,方鸿生.用反向传播人工神经网络预测低碳低合金钢的马氏体转变开始温度[J].金属学报,2003,39(6):630-634.