摘要
针对Q235B热轧带钢性能预测系统,提出一种回归分析和神经网络相结合的方法来预测其力学性能.首先,测量材料最终相的组成与铁素体的晶粒度,应用多重回归分析的方法,建立成分、相体积分数、晶粒尺寸与抗拉强度、屈服强度、延伸率的对应关系.另一方面,采用BP神经网络方法,结合相变动力学模型的计算数据,通过大量数据的自学习训练,完成神经网络模型对抗拉强度、屈服强度、延伸率的预测.预测结果表明,应用神经网络和回归分析方法,具有较高的预测精度.
In order to develop on-line or off-line predictive system of hot rolling process, a model combining the neural network and regression methods was proposed to predict the properties of hot rolled plain carbon steel Q235B. On the base of the measured data to the grain size of ferrite and the fraction of phases at room temperature, a multiple regression method is given to describe the relationship between mechanical property and chemical composition, grain size, fraction of phase and so on. In the present work, the error back-propagation network (BP) is adopted. Combined with the data calculated with the phase transformation kinetic model, a great deal of data were trained many times and compared with the experimental data to these steels, the BP model predicting yield strength, tensile strength, elongation was built. The results show that the properties of hot rolling strip can be precisely predicted using neural network and regression method.
出处
《金属学报》
SCIE
EI
CAS
CSCD
北大核心
2003年第10期1110-1114,共5页
Acta Metallurgica Sinica
基金
国家高技术研究发展计划项目2001AA339030
沈阳工业学院基金项目No.3200903资助
关键词
热轧带钢
神经网络
回归
hot rolling strip
neural network
regression
