摘要
为实现超声辅助滚挤压工艺参数的优化控制,设计了以工艺参数(转速、进给速度、振幅、静压力)为影响因素、以表层性能(表面粗糙度、残余应力、硬度)为响应值的4因素5水平正交试验。基于试验数据构建了RBF神经网络预测模型,并进行了准确性检验,解决了传统BP神经网络易陷入局部最优的问题;基于SPEA2SDE与NSGA-Ⅱ算法进行了二维双目标ZDT测试函数和三维多目标DTLZ测试函数的性能测试,并通过对比两种算法的Pareto前沿解证明了SPEA2SDE算法的优异性。最后进行了超声辅助滚挤压试验验证,优化后的表层性能最大误差均控制在5%以下,证明了优化算法的精确性。
To optimize the process parameters for ultrasonic-assisted roll extrusion,a four-factor and five-level orthogonal experiment was designed by taking the process parameters(rotational speed,feeding rate,amplitude and static pressure)as the influence factors,and taking the surface properties(surface roughness,residual stress and hardness)as the response values.Based on the experimental data,a predictive model was established using a Radial Basis Function(RBF)neural network,and its accuracy was validated.This approach addresses the issue of the traditional Backpropagation(BP)neural networks being prone to converging to local optima.Performance tests were carried out on the two-dimensional bi-objective ZDT and three-dimensional multi-objective DTLZ test functions using the SPEA2SDE and NSGA-II algorithms.The Pareto front plots generated by the two algorithms demonstrated the superiority of the SPEA2SDE algorithm.Finally,ultrasonic-assisted roll extrusion experiments were conducted for validation.The maximum error between the predicted and measured surface properties was less than 5%,confirming the accuracy of the optimization algorithm.
作者
刘玲玲
付浩然
Liu Lingling;Fu Haoran(School of Intelligent Engineering,Zhengzhou College of Finance and Economics,Zhengzhou 450000,China;Zhengzhou Key Laboratory of Intelligent Assembly Manufacturing and Logistics Optimization,Zhengzhou 450000,China;College of Mechanical and Electrical Engineering,Henan University of Science and Technology,Luoyang 471003,China)
出处
《锻压技术》
北大核心
2026年第2期161-170,共10页
Forging & Stamping Technology
基金
河南省科技攻关项目(222102210202,242102221005)
河南省本科高校青年骨干教师培养计划(2024GGJS179)。
