Design of hot rolling process based on Bayesian neural network and multi-objective optimization
In this paper,an integrated framework including the prediction of mechanical properties and design of hot rolling process to achieve the required mechanical properties was established.The Bayesian neural network was employed to develop the relationship between chemical composition,process parameters and mechanical properties due to its excellent generalization ability,with prediction precisions of ±6%,±6% and ±4% for yield strength,tensile strength and elongation,respectively.An error bar was indicated on each prediction value to index the prediction reliability.The adaptive weighted PSO (AWPSO)algorithm based on multi-objective optimization was developed to locate the Pareto front as the processing windows.The design of hot rolling process was performed with the required mechanical properties or the optimal combination of mechanical properties and processing routes set to be the optimization targets.Good agreements have been achieved between optimization results and industrial trials,which could provide practical guidance for the design of hot rolling processes.
Bayesian neural network adaptive weighted PSO process optimization hot rolling
Jia Tao Liu Zhenyu Liu Xianghua Wang Guodong
The State Key Laboratory of Rolling & Automation ,Northeastern University,Shenyang 110004,China
国际会议
上海
英文
1007-1011
2008-09-26（万方平台首次上网日期，不代表论文的发表时间）