基于神经网络的铝合金机械圆盘冲锻工艺优化  

Optimization of Stamping-Forging Process for Aluminum Alloy Mechanical Disk Based on Neural Network

作　　者：雷艳惠[1] 赵芳霞[2] LEI Yanhui;ZHAO Fangxia(Xianyang Vocational&Technical College,Xianyang 712000,China;College of Material Science and Engineering,Nanjing Tech University,Nanjing 210009,China)

机构地区：[1]咸阳职业技术学院,陕西咸阳712000 [2]南京工业大学材料科学与工程学院,江苏南京210009

出　　处：《热加工工艺》2021年第5期102-105,共4页Hot Working Technology

基　　金：2017年度咸阳职业技术学院科研项目(2017KYB05)。

摘　　要：采用7×42×1三层拓扑结构构建了铝合金机械圆盘冲锻工艺神经网络优化模型,并进行了训练、预测验证和产线应用。结果表明,该模型输出的磨损体积相对训练误差范围为3.14%~5.00%,平均相对训练误差为4.11%;相对预测误差范围为3.71%~5.52%,平均相对预测误差为4.31%。应用该优化模型成形的铝合金机械圆盘的磨损体积比产线现用工艺减小了24.24%,耐磨损性能显著提高。该神经网络优化模型预测性准、精度高。铝合金机械圆盘冲锻工艺的最佳参数为:摩擦系数0.15、凸模速度15 mm/s、冲裁力8 kN、模具预热温度250℃、反顶力6.5 kN、压边力5.5 kN、锻造温度480℃。The neural network optimization model of aluminum alloy mechanical disc stamping-forging process was constructed with a 3-layer topological structure of 7 ×42 ×1. And its training, prediction, verification and production line application were carried out. The results show that the relative training error range of wear volume output by this model is 3.14%-5.00%, and the average relative training error is 4.11%. The range of relative prediction error is 3.71%-5.52%, and the average relative prediction error is 4.31%. The wear volume of the aluminum alloy mechanical disk formed by this optimization model is24.24% less than that of the current production line, and the wear resistance is significantly improved. The neural network optimization model has accurate predictability and high precision. The optimum parameters of punching-forging process for aluminum alloy mechanical disk are friction coefficient of 0.15, punch speed of 15 mm/s, blanking force of 8 kN, die preheating temperature of 250 C, counterforce of 6.5 kN, blank holder force of 5.5 kN and forging temperature of 480℃.

关 键 词：神经网络 铝合金 机械圆盘 冲锻工艺 磨损性能 

分 类 号：TG306[金属学及工艺—金属压力加工]