基于神经网络航空发动机曲轴加工工艺设计优化  

Design and optimization on machining process for aircraft engine crankshaft based on neural network

作　　者：肖展开 梅益[1] 罗宁康 袁丽莉 张然 吴巧 Xiao Zhankai;Mei Yi;Luo Ningkang;Yuan Lili;Zhang Ran;Wu Qiao(School of Mechanical Engineering,Guizhou University,Guiyang 550025,China;School of Big Data and Information Engineering,Guizhou University,Guiyang 550025,China)

机构地区：[1]贵州大学机械工程学院,贵州贵阳550025 [2]贵州大学大数据与信息工程学院,贵州贵阳550025

出　　处：《锻压技术》2022年第6期35-46,86,共13页Forging & Stamping Technology

基　　金：贵州省科技计划项目(黔科合支撑[2019]2019);贵州省科技支撑计划项目(黔科合支撑[2018]2175)。

摘　　要：以某发动机曲轴作为研究对象,通过DEFORM软件对曲轴预锻和终锻过程进行了金属流动和微观组织的耦合模拟,得出了不同工艺参数下曲轴锻造的模拟数值;采用Geomagic软件逆向工程技术,得到了最终处理完成后曲轴的实际尺寸,获得体积验证参数;最后,通过FORGE软件验证了其结果的正确性。建立了多输入、多输出的BP网络拓扑结构,以预锻温度、终锻温度、预锻阻尼墙高度、终锻阻尼墙高度、有无润滑、预锻速度和终锻速度等作为神经网络模型的输入参数,以终锻填充率、折叠量以及最大应力值性能指标作为神经网络模型的输出参数。通过神经网络预测值与正交试验下模拟值的对比,得出最优解下的工艺参数,发现训练后BP神经网络模型的相对误差较小,并具有很好的预测能力。For an engine crankshaft, the coupled simulation of metal flow and microstructure was carried out on crankshaft pre-forging and final forging processes by DEFORM software, and the simulation values of crankshaft forging under different process parameters were obtained. Then, the actual dimension of crankshaft after final treatment was obtained by Geomagic software reverse engineering technology, the volume verification parameters were obtained, and the correctness of the results was verified by FORGE software. Furthermore, a multi-input and multi-output BP network topology was established taking pre-forging temperature, final forging temperature, pre-forging damping wall height, final forging damping wall height, with or without lubrication, pre-forging speed and final forging speed as input parameters of neural network model and performance indexes of final forging filling rate, folding amount and maximum stress value as output parameters of neural network model. Finally, the optimal process parameters were obtained by comparing the predicted values by neural network with simulated values under orthogonal test. The results show that the relative error of the trained BP neural network model is small and has good predictive ability.

关 键 词：神经网络 曲轴 锻造 逆向工程 正交试验 金属流动 微观组织 

分 类 号：TG316.3[金属学及工艺—金属压力加工]