基于BP网络的Al-Cu-Mg-Sc-Zr合金均匀化电导率预测

来源期刊：中南大学学报(自然科学版)2010年第5期

论文作者：邓英 尹志民 侯延辉 王华

文章页码：1761 - 1766

关键词：Al-Cu-Mg-Sc-Zr合金；均匀化；电导率；BP神经网络；Levenberg-Marquardt算法

Key words：Al-Cu-Mg-Sc-Zr alloy; homogenization; electro-conductivity; BP neural network; Levenberg-Marquardt algorithm

摘    要：采用电导率测量、X线衍射物相分析研究不同均匀化处理条件下Al-Cu-Mg-Sc-Zr合金的电导率和组织变化规律。研究结果表明：Al-Cu-Mg-Sc-Zr铸态合金中固溶体过饱和程度较高，电导率较低；在360 ℃以下均匀化，亚稳的过饱和固溶体析出平衡相θ和S，基体固溶体过饱和程度下降，电导率上升；在360 ℃以上均匀化，平衡相逐渐回溶入固溶体基体，基体固溶体过饱和程度上升，电导率下降。在此基础上，采用Levenberg-Marquardt算法对权值和阈值进行训练，建立以均匀化热处理温度和时间与电导率高度非线性BP神经网络预测模型。该模型能较好地反映均匀化工艺参数与电导率之间的内在规律，泛化检测点电导率相对误差为-0.006~0.006，表明该神经网络模型具有较高的预测精度和较强的泛化能力。

Abstract: The electrical conductivity and the microstructure of Al-Cu-Mg-Sc-Zr alloy under various homogenization conditions were investigated by electrical conductivity measurement and X-ray diffraction. The results show that the higher degree of super saturation in the as-cast alloy leads to the lower electrical conductivity; when homogenization temperature is below 360 ℃, the super saturation solid solution of as-cast alloy is decomposed and θ and S equilibrium phases precipitates, the degree of super saturation in the matrixes decreases and the electrical conductivity increases; when homogenization temperature is above 360 ℃, equilibrium phases are dissolved into matrix and the electrical conductivity decreases again. Based on the research, a predicting model was established about the relationship between the homogenization conditions and the electro-conductivity. The model was based on error back-propagation (BP) algorithm and trained by Levenberg-Marquardt algorithm. The model reflects the internal law between the process parameters and electrical conductivity better, and the relative error of generalization is not more than -0.006-0.006, which indicates the model has high precision and good generalization performance.

基金信息：国家“973”重点基础研究项目