BP神经网络预测冲击强化45钢的中温热稳定性
Prediction of Medium Thermal Stability of 45 Steel after Impact Strengthening with BP Artificial Neural Network
姬帅 1张佳乐 2王海丽2
作者信息
- 1. 西安石油大学 材料科学与工程学院,陕西 西安 710065;南京工业大学 材料化学工程国家重点实验室,江苏 南京 210009
- 2. 西安石油大学 材料科学与工程学院,陕西 西安 710065
- 折叠
摘要
室温下采用自由落体式对正火态 45 钢进行冲击强化,对冲击强化正火态 45 钢进行中温时效处理,分别加热至 450、550、650℃,每组温度均保温 10、20、30、40 min,同时对各组试样进行显微硬度测试,并对加热至 650℃的 4种试样进行显微组织观察;以试样的实际状态参量作为学习样本对 3 层BP神经网络进行训练.结果表明:BPANN能够对冲击强化正火态 45 钢的中温热稳定性进行预测,且误差可以控制在 3%～6%;BPANN的预测值均大于实测值,但是预测值的变化趋势与实测值的变化趋势一致,网络的预测精度可以通过提高误差函数的收敛速率来得到提高.通过对 650℃试样显微组织的观察,可以判定网络的输入层涉及的相关内容能让BPANN的预测结果反映出材料的真实状态.本研究可以降低实验成本、减少实验数量,有助于对冲击强化正火态 45 钢在其他加热温度下的热稳定性进行预测.
Abstract
The impact strengthened normalized 45 steel which was impact strengthened with free-fall type at room temperature have been aged at medium temperature.The steel was heated to 450℃,550℃and 650℃,respectively.Temperature of each group was kept for 10 min,20 min,30 min and 40 min,and the microhardness of each group was tested.Micro-structure of four kinds of samples heated to 650℃was observed.Taking the actual state parameters of samples as the learning sample,the three-layer BP neural network was trained.The results show that BPANN can predict the thermal stability of impact strengthened normalized 45 steel at medium temperature,and the error can be controlled within 3%-6%.Predicted values of BPANN are all larger than the measured ones,but the variation trend of predicted values is consistent with that of measured values.Prediction accuracy of the network can be improved by increasing the convergence rate of error function.Through observation on microstructure of the sample at 650℃,it can be determined that relevant contents involved in input layer of the network can make prediction result of BPANN reflect real state of the material.This work can reduce experimental cost and number of experiments,it's helpful to predict the thermal stability of impact strengthened normalized 45 steel at other heating temperatures.
关键词
人工神经网络/BP算法/冲击强化/正火态45钢/中温热稳定性Key words
artificial neural networks/BP method/impact strengthening/normalized 45 steel/medium-temperature ther-mal stability引用本文复制引用
出版年
2024
NETL
NSTL
万方数据