摘要
为确定航空发动机叶片机器人磨抛过程中材料去除深度与工艺参数之间的关系,获得加工所需的工艺参数,实现叶片表面材料的定点定量去除,建立叶片机器人磨抛加工系统,将各工艺参数考虑在内进行多组正交实验;利用实验数据建立基于最小二乘支持向量回归机(least squares support vector regression,LSSVR)模型,利用增强型鲸鱼优化算法(enhanced whale optimization algorithm,EWOA)提高算法精度、寻优能力和避免陷入局部最优并对LSSVR的超参数进行优化;对比标准鲸鱼优化算法(whale optimization algorithm,WOA)和粒子群优化(particle swarm optimization,PSO)算法预测模型的结果,并利用模型预测的工艺参数进行实验验证。结果表明:EWOA-LSSVR预测模型的决定系数R为96.031%,平均绝对误差RMAE为0.012128 mm,相较于WOA-LSSVR和PSO-LSSVR模型具有更好的拟合度;且验证实验结果证明EWOA-LSSVR预测模型具有较好的预测准确性,并可为叶片表面材料的定点定量去除提供可靠依据。
Objectives:High-pressure turbine blades,as the core components of aviation engines,are subjected to harsh working environments of high temperature,high pressure,and high load for a long time,which places strict requirements on their high-temperature mechanical properties and structural stability.Therefore,the material of turbine blades is often selected as single-crystal high-temperature alloys and the blades are made through precision casting processes.Due to the casting characteristics of the blades,the material distribution of the workpiece is uneven,that is,the deviations of the design sizes from different positions on the blade surfaces vary.Therefore,the fixed-point quantitative removal of the blade surface material plays a very important role in the blade production and manufacturing process.Methods:Blade grinding and polishing processing experiments are established by considering various technological parameters.The experimental data are used as the training set for the prediction model,and a prediction model based on the least squares support vector machine(LSSVR)is constructed.In the LSSVR hyperparameter setting stage,the enhanced whale optimization algorithm(EWOA)is used to improve algorithm accuracy,enhance optimization capability,and prevent local optima while optimizing the LSSVR hyperparameters.The prediction models optimized by other algorithms are established for comparison of model prediction capabilities.The prediction results are applied to the reproduction experiments of the material removal amount,and the performance of the prediction model is evaluated by using the processing results.Results:From the perspective of model establishment and result prediction,the processing parameter prediction model EWOA-LSSVR based on the enhanced whale optimization algorithm(EWOA)-optimized least squares support vector machine(LSSVR)exhibits high prediction accuracy and good model fitting degree,with a determination coefficient of 96.031%and a mean absolute error RMAE of 0.012128 mm.The prediction models of LSSVR optimized by the whale optimization algorithm(WOA)and particle swarm optimization(PSO)have determination coefficients of 89.457%and 92.228%,and mean absolute errors(RMAE)of 0.012358 and 0.012462 mm,respectively.In contrast,the prediction results of EWOA-LSSVR are more accurate with lower errors.The prediction results of EWOALSSVR are used as the process parameters for blade processing.When the dimensional error of the processed area of the blade enters the design tolerance zone of±0.05 mm,it is considered qualified.The qualified rate of the sampling points in the two processing experiments reaches 93.59%,which plays a certain guiding role in the actual processing of the blade.Conclusions:A prediction model for process parameters is established by using the least squares support vector machine suitable for small sample sizes.To improve the algorithm accuracy of model establishment and avoid falling into local optima,the enhanced whale algorithm is adopted to optimize the hyperparameters of the least squares support vector machine,and a prediction model with a determination coefficient of 96.031%and an average absolute error of 0.012128 mm is established.By comparing with the prediction models optimized by WOA and PSO,the established prediction model has certain advantages in terms of determination coefficient,mean absolute error and mean square error.The reproduction experiment of the removal amount is carried out.After two processing experiments,a processing result with a qualified rate of 93.59%at the sampling points is achieved,proving the feasibility of using this method to achieve fixed-point and quantitative removal of the blade surface material.
作者
张诗涵
魏锦辉
王阳
朱光
李论
刘殿海
ZHANG Shihan;WEI Jinhui;WANG Yang;ZHU Guang;LI Lun;LIU Dianhai(Shenyang Institute of Automation,Chinese Academy of Sciences,Shenyang 110016,China;School of Mechanical Engineering&Automation,University of Science and Technology Liaoning,Anshan 114051,Liaoning,China)
出处
《金刚石与磨料磨具工程》
北大核心
2025年第4期551-560,共10页
Diamond & Abrasives Engineering
基金
辽宁省自然科学基金(2023-MS-034)
研究所基础研究面上项目(20222JK2K09)
国家资助博士后研究人员计划(GZC20232882)
中国博士后面上科学基金(2023M743703)。
关键词
机器人砂带磨抛
工艺参数
机器学习
最小二乘支持向量回归机
增强型鲸鱼优化算法
robot abrasive belt grinding and polishing
process parameter
machine learning
least squares support vector regression machine(LSSVR)
enhanced whale optimization algorithm(EWOA)
