基于神经网络-遗传算法的泵体零件热锻模具磨损与应力分析  

Analysis on wear and stress of hot forging die for pump body parts based on neural network-genetic algorithm

作　　者：刘寿军 肖开永 牛腾 王永 王雷刚[1] Liu Shoujun;Xiao Kaiyong;Niu Teng;Wang Yong;Wang Leigang(College of Materials Science and Engineering,Jiangsu University,Zhenjiang 212013,China;Kunshan Huizhong Electromechanical Co.,Ltd.,Kunshan 215331,China)

机构地区：[1]江苏大学材料科学与工程学院,江苏镇江212013 [2]昆山惠众机电有限公司,江苏昆山215331

出　　处：《锻压技术》2024年第2期208-214,共7页Forging & Stamping Technology

基　　金：国家自然科学基金资助项目(51775249)。

摘　　要：基于Archard磨损模型和BP神经网络-遗传算法,采用有限元模拟研究泵体零件锻造过程中工艺参数对模具的磨损深度及最大应力的影响规律。首先,选取锻造温度、预热温度、摩擦因数、硬度和下压速度进行5因素5水平正交试验,借助灰色关联的分析确定关键影响因子,即锻造温度和摩擦因数。然后,采用拉丁超立方抽样法对关键影响因子随机取样,并采用DEFORM软件进行模拟,再将锻造温度和摩擦因数作为输入,将磨损深度和最大应力作为输出,构建2×12×2的神经网络-遗传算法优化模型,确定最优锻造温度为481.6℃、摩擦因数为0.38,此时最小磨损深度为1.44×10^(-5) mm、最大应力为1362 MPa。最后,通过实际生产进行验证,得到磨损深度和最大应力的模拟值与预测值吻合,相对误差仅为0.7%和4.0%,主要磨损区域和应力分布情况与实际生产基本吻合,验证了优化模型的可行性。Based on Archard wear model and BP neural network-genetic algorithm,the influence laws of process parameters on die wear depth and maximum stress during the forging process of pump body parts were studied by finite element simulation.Firstly,the orthogonal test of five factors and five levels was conducted by selecting forging temperature,preheating temperature,friction coeicient,hardness and pressing speed,and the key influencing factors,namely forging temperature and friction factor were determined by grey correlation a-nalysis.Then,the key influencing factors were randomly sampled by Latin hypercube sampling method,and the simulation was conducted by software DEFORM.Furthermore,taking forging temperature and friction coefficient as the input and wear depth and maximum stress as the output,a 2×12×2 neural network-genetic algorithm optimization model was constructed,and the optimum process parameters were de-termined as the forging temperature of 481.6℃,the friction coefficient of 0.38,the minimum wear depth of 1.44×10^(-5) mm and the maximum stress of 1362 MPa.Finally,through actual production verification,the simulated values of wear depth and maximum stress were consistent with the predicted values,with the relative errors of only 0.7%and 4.0%,and the main wear area and the stress distribu-tion were basically consistent with the actual production,verifying the feasibility of the optimized model.

关 键 词：神经网络 遗传算法 Archard磨损模型 锻造温度 模具磨损深度 

分 类 号：TG316.3[金属学及工艺—金属压力加工]