摘要
为有效缩短7003铝合金热处理工艺,采用误差回转(BP)神经网络控制的方案,利用梯度下降法导出网络算法,建立热处理工艺与硬度之间BP神经网络模型。结果表明:预测值和实测值吻合较好,克服了以往单因素实验法不能正确有效反映热处理工艺和硬度之间高度非线性、时变性的弱点。该方法为有效、快捷、经济地开发有色金属和黑色金属的热处理工艺优化提供了新的思路。
In order to shorten the fussy experimental process in heat treatment of 7003 aluminum alloy, backpropagation neural network control of scheme was proposed. By using gradient descent algorithms, a network of arithmetic was induced. Between heat treatment technics and the hardness BP neural network was set up. The results indi- cate that predicted and test results were identical and thus the weakness of nonlinear and time-variation relationship beween heat treatment and hardness caused by using single-factor-experiment method was overcome. The proposed method provides a new thinking to develop heat treatment optimization of metals effectively, quickly and economically.
出处
《宇航材料工艺》
CAS
CSCD
北大核心
2009年第4期6-9,57,共5页
Aerospace Materials & Technology
基金
国家自然科学基金资助项目(50771093)
关键词
误差回转神经网络
梯度下降算法
7003铝合金
热处理
Back-propagation neural networks, Gradient descent algorithms ,7003 aluminum alloy, Heat treatment
