基于神经网络的RE-Ni-Cu合金铸铁腐蚀性能预测

通过动态质量损失法腐蚀试验获取RE-Ni-Cu合金铸铁在高温浓碱液中的实测腐蚀深度,并将其作为样本数据用于BP神经网络的训练和验证;利用MATLAB的工具箱函数分别建立拓扑结构为4×15×1和4×15×8×1的BP神经网络,并对两个网络模型进行比较研究。结果表明,在样本集和训练条件下,4层BP网络的预测精度明显高于3层BP网络,可用于RE-Ni-Cu合金铸铁在高温浓碱液腐蚀系统中的腐蚀性能预测。

Corrosion prediction of RE-Ni-Cu alloy cast iron based on neural network

The corrosion rate of Re-Ni-Cu alloy cast iron in high temperature concentrated alkaline solution were measured by the dynamic condition hydrometer method, and it was the sample data used for training and certification of BP neural network.The 4×15×1 and 4×15×8×1 BP neural network models were established by the toolbox function of MATLAB, and two network models were comparatively researched.The results show that under this sample set and training condition, prediction precision of four layer BP neural network is obviously higher than that of three layers BP neural network, and four layers BP neural network canbe used for corrosion prediction of RE-Ni-Cu alloy cast iron in high temperature concentrated alkaline solution.