基于PSO-BP神经网络模型的654SMO热变形行为预测
Prediction of thermal deformation behavior of 654SMO based on PSO-BP neural network model
张博文 1闫德安 1和鹏越 1管煜 1陈锐 1刘元铭2
作者信息
- 1. 太原理工大学机械与运载工程学院,山西 太原 030024
- 2. 太原理工大学机械与运载工程学院,山西 太原 030024;太原理工大学先进金属复合材料成形技术与装备教育部工程研究中心,山西 太原 030024;海安太原理工大学先进制造与智能装备产业研究院,江苏 海安 226601
- 折叠
摘要
针对 654SMO 超级奥氏体不锈钢的热变形行为进行研究,采用 Arrhenius 模型与粒子群算法优化的 BP神经网络模型(PSO-BP神经网络)对 654SMO 超级奥氏体不锈钢热变形行为进行预测,将其结果进行对比获得最优模型.通过实验获得变形温度在 1 000～1 200℃、应变速率为 0.1～10 s-1 条件下的真应力,并采用考虑应变修正的 Arrhenius模型和PSO-BP神经网络模型对实验数据进行训练,通过计算均方相关系数(R2)、均方根误差(RMSE)和平均相对误差(AARE),对预测结果量化并且进行比较.最后基于实验实测数据和 PSO-BP 模型的预测数据得到应变为 0.3 和 0.6 的热加工图.结果表明:相比于传统的 Arrhenius 模型,PSO-BP 神经网络模型具有更高的准确性和适用性,为 654SMO 的热加工工艺提供理论指导.
Abstract
The hot forming behavior of 654SMO super austenitic stainless steel was researched.The Arrhenius model and PSO-BP neural network model were used to predict the hot deformation behavior of 654SMO super austenitic stainless steel,and the results were compared to select the optimal model.In addition,the deformation stress was obtained through experiments at temperatures ranging from 1 000℃to 1 200℃and strain rates ranging from 0.1 to 10 s-1.The Arrhenius model considering strain correction and the PSO-BP neural network model were used to train the experimental data.The predicted results were quantified and compared by calculating the mean square correlation coefficient(R2),mean square error(RMSE)and average relative error(AARE).Finally,hot working diagrams with strain of 0.3 and 0.6 were created based on the current experimental data and the predicted data of the PSO-BP model.The prediction results show that the constructed PSO-BP neural network has a higher accuracy and applicability than the Arrhenius model and can provide theoretical clues for the hot working process of 654SMO.
关键词
654SMO/超级奥氏体不锈钢/热变形预测/Arrhenius本构方程/PSO-BP神经网络Key words
654SMO super austenitic stainless steel/thermal deformation prediction/Arrhenius constitutive equa-tion/PSO-BP neural network引用本文复制引用
基金项目
国家重点研发计划(2021 YFB3401000)
国家级大学生创新训练项目(202210112021)
国家自然科学基金(52375367)
国家自然科学基金(51904206)
海安太原理工大学先进制造与智能装备产业研究院开放研发项目(2023 HA-TYUTKFYF028)
出版年
2024
NETL
NSTL
万方数据