基于机器学习的TC18钛合金起落架锻造成形工艺智能优化  

Intelligent optimization of forging forming process of TC18 titanium alloy landing gear based on machine learning

作　　者：唐学峰 王志洲 冯仪 余俊 邓磊[1] 金俊松[1] 王新云[1] TANG Xue-feng;WANG Zhi-zhou;FENG Yi;YU Jun;DENG Lei;JIN Jun-song;WANG Xin-yun(State Key Laboratory of Materials Processing and Die&Mould Technology,Huazhong University of Science and Technology,Wuhan 430074,China;Wuhan NEWWISH Technology Co.,Ltd.,Ezhou 436070,China)

机构地区：[1]华中科技大学材料成形与模具技术全国重点实验室,湖北武汉430074 [2]武汉新威奇科技有限公司,湖北鄂州436070

出　　处：《塑性工程学报》2024年第4期254-266,共13页Journal of Plasticity Engineering

基　　金：国家自然科学基金资助项目(52105337,52090043);国家重点研发计划(2022YFB3706903)。

摘　　要：针对TC18钛合金加工窗口窄、零件成形质量与性能的预测以及工艺参数的优化设计困难的问题,通过热压缩试验分析了TC18钛合金的热变形行为与微观组织演变规律。基于Bayesian算法优化构建了TC18钛合金热变形流动应力和微观组织演变的深度神经网络(DNN)模型。通过对UG和Deform的二次开发,完成了不同工艺参数下起落架锻造的自动建模与仿真,并建立了预锻件的尺寸-工艺-质量数据库。结合DNN和遗传算法(GA)、粒子群算法(PSO)、快速非支配排序遗传算法(NSGA2)确定了最优的预锻件工艺参数。结果表明,采用NSGA2优化后锻件最大成形力R_(Tmax)可降低40.6%,目标截面的平均初生α相含量为0.207,接近最优含量20.0%。In view of the problems of narrow processing window of TC18 titanlum alloy,and the difficulties in predicting the forming quality and performance of parts and optimizing the design of process parameters,the thermal deformation behaviors and microstructure evolution laws of TC18 titanium alloy were analyzed by hot compression expenments.The deep neural network(DNN)model for the thermal deformation flow stress and microstructure evolution of TC18 titanium alloy was constructed based on Bayesian algorithm optimization.Through the secondary development of UG and Deform,the automatic modeling and simulation of landing gear forging with different process parameters were completed,and the size-process-quality database of preforging parts was established.Combined with DNN,genetic algorithm(GA),particle swarm optimization(PSO)and fast non-dominated sorting genetic algorithm(NSGA2),the optimal preforging process parameters were determined.The results show that the maximum forming force of forging is reduced by 40.6%after NSGA2 optimization,and the content of primaryαphase in the target cross-section is 0.207,which is close to the optimal content of 20.0%.

关 键 词：TC18钛合金 深度神经网络 智能优化算法 本构模型 微观组织模型 起落架锻造 

分 类 号：TG316.8[金属学及工艺—金属压力加工]