基于PNN法的Zr-4合金管材矫直参数与氢化物取向因子的研究

Modeling of Hydride Orientation Factor in Zr-4 Alloy Tube on Straightening of the Parameters by PNN

本文基于PNN的算法,建立了矫直参数(入口矫直辊角度、中间矫直辊角度、出口矫直辊角度、压下量及弯曲度)与Zr-4合金管材氢化物取向因子的网络预测模型。输入层为入口矫直辊角度、中间矫直辊角度、出口矫直辊角度、压下量和弯曲度,输出层为各矫直参数下的Zr-4合金管材的氢化物取向因子F40。网络模型结构为3-0.6-1,均方误差最小值为0.012、网络敏感性分析认为入口矫直辊角度,中间矫直辊角度、出口矫直辊角度对网络的精度影响较小,而压下量和弯曲度对网络精度影响很大。随着矫直压下量和弯曲度的增加,Zr-4合金氢化物取向因子增加。基于PNN方法构建的Zr-4合金管材的矫直参数与氢化物取向因子的网络模型可以有效的根据矫直参数预测Zr-4合金氢化物取向因子结果。

In this paper, the probabilistic neural networks（PNN） module was used to predict the hydride orientation factor F40 after straightening in Zr-4 alloy tube. In this module, the straightening parameter（the angle of entrance roll, the angle of middle roll, the angle of export roll, reduction and bending） as input units while the hydride orientation was employed as output unit. The results show that the optimal network architecture is considered to be 3-0.6-1, and the root mean square erro（rRMSE）of the training data is 0.012.The sensitivity analysis（SA） show that reduction and bending are the most important parameter for the PNN model accuracy, the entrance of straightening roll angle, the middle of straightening roll angle, the export of straightening roll Angle are less important. Higher reduction and bending are leading to increase of the hydride orientation factor F40 in Zr-4 alloy tube. The PNN model is an efficient tool to evaluate and predict the hydride orientation factor F40 in Zr-4 alloy tube under different straightening parameters.