摘要
[目的]数控机床在运行状态下具有瞬时功率高及能源效率低的特点,并且数控机床的加工能耗随着复杂多变的加工任务实时变化,导致机床加工能耗预测困难。基于信息流与能量流的数控机床加工能耗预测机理模型需要操作人员能辨悉机床运行状态,明悉机床能耗变化特点,故机床加工能耗预测困难且周期较长。随着测试技术与计算机算力的显著提升,基于数据驱动的预测方法被引入数控机床加工能耗预测研究中。因此,本文提出一种结合随机配置网络(SCNs)与多机制改进后沙丘猫(SCSO)算法的自适应增量式机器学习方式,实现数控机床加工能耗的高效、高精预测。[方法]以螺旋槽数控铣床铣削螺杆转子为例,基于工艺参数设计螺旋槽数控铣床加工能耗铣削实验,采集加工能耗数据。采用多机制改进后的沙丘猫算法优化随机配置网络,建立数控机床加工能耗预测模型。将随机配置网络算法作为机床加工能耗的预测模型,结合Tent种群初始化策略、可变螺旋搜索策略与自适应t分布策略改进后的沙丘猫算法对随机配置网络建模过程中的比例因子与正则化参数进行求解,提升随机配置网络的预测精度及预测效率。[结果]为验证模型的准确性,以均方根误差(RMSE)和平均绝对百分比误差(MAPE)为评价指标,将SCSO-SCNs算法、SCNs算法、松鼠算法优化的BP神经网络(SSA-BP)进行对比,结果表明,SCSO-SCNs算法较SSA-BP算法与SCNs算法在RMSE上分别减少了38.62%与46.03%,同时在MAPE上较SSA-BP算法与SCNs算法分别减少了40.47%与47.33%,证明SCSO-SCNs模型在加工能耗预测方面性能的优越性。[结论]提出的融合随机配置网络与多机制改进的沙丘猫算法的机器学习方法在数控机床加工能耗预测领域展现出较为明显的性能优势。改进后的沙丘猫算法通过优化算法初始种群、改进种群位置更新策略,提升了算法搜索效率及跳出局部最优解能力。基于多机制改进后沙丘猫算法寻求随机配置网络最佳比例因子与泛化因子,极大地提升了模型的预测精度,经过与现有机器学习算法比较可知,本文提出的方法预测精度更高,且极大地提升了机床加工能耗的预测效率。
[Objective]CNC machine tools in the operating state feature high instantaneous power and low energy efficiency,and their machining energy consumption power changes with complex and variable processing tasks in real time,thus making it difficult to predict energy consumption of machine tool machining.The prediction mechanism model of machining energy consumption of CNC machine tools based on information flow and energy flow requires operators to be aware of the operating status of the machine tool and the characteristics of energy consumption changes of the machine tool,which results in difficult prediction of machining energy consumption of machine tools and long cycles.As the testing techniques and computational power of computers significantly improve,data-driven prediction methods have been introduced to the research on predicting the machining energy consumption of machine tools.Therefore,an adaptive incremental machine learning approach that combines stochastic configuration networks(SCNs)with a multi-mechanism-improved sand cat swarm optimization(SCSO)was proposed to achieve efficient and high-precision prediction of the machining energy consumption of machine tools.[Methods]By taking the helical groove CNC milling machine milling screw rotor as an example,based on the process parameters,the machining energy consumption milling experiments and collected machining energy consumption data were designed.Meanwhile,SCNs were optimized by adopting the multi-mechanism-improved SCSO algorithm to build a prediction model for machining energy consumption.The SCN algorithm was employed as the prediction model for machining energy consumption.The SCSO algorithm improved by combining the Tent population initialization strategy,variable helix search strategy and adaptive t-distribution strategy solved the scale factor and regularization parameter during SCN modeling to improve the prediction accuracy and prediction efficiency of SCNs.[Results]To verify the accuracy of the model,the root mean square error(RMSE)and mean absolute percentage error(MAPE)were employed as the evaluation indexes to compare the BP neural networks(SSA-BP)optimized by the SCSO-SCNs,SCNs,and squirrel search algorithm.Comparison results show that compared to SSA-BP and SCNs,SCSO-SCNs shows a decrease of 38.62%and 46.03%in RMSE respectively,while it presents a reduction of 40.47%and 47.33%in MAPE compared to SSA-BP and SCNs respectively,which proves the performance superiority of the SCSO-SCNs model in the prediction of machining energy consumption.[Conclusions]The proposed machine learning method,which integrates SCNs and multi-mechanism-improved SCSO algorithm,shows more obvious performance advantages in terms of machining energy consumption prediction for CNC machine tools.The improved SCSO algorithm enhances the search efficiency and the ability to jump out of local optimal solutions by optimizing the initial population of the algorithm and improving the population position updating strategy in the improvement and exploitation phases.The SCSO algorithm,based on multi-mechanism improvement,greatly improves the prediction accuracy of the model by seeking the optimal scale factor and generalization factor of SCNs.Comparison with existing machine learning algorithms shows that the proposed method has higher prediction accuracy and greatly improves the prediction efficiency of machining energy consumption.
作者
张维锋
孙兴伟
刘寅
赵泓荀
穆士博
ZHANG Weifeng;SUN Xingwei;LIU Yin;ZHAO Hongxun;MU Shibo(School of Mechanical Engineering,Shenyang University of Technology,Shenyang 110870,Liaoning,China;Key Laboratory of Numerical Control Manufacturing Technology for Complex Surfaces of Liaoning Province,Shenyang University of Technology,Shenyang 110870,Liaoning,China)
出处
《沈阳工业大学学报》
北大核心
2025年第6期775-782,共8页
Journal of Shenyang University of Technology
基金
辽宁省应用基础研究计划项目(2022JH2/101300214)
辽宁省教育厅面上项目(JYTMS20231190,LJKMZ20220459)
辽宁省教育厅科技创新团队项目(222410142011)。
关键词
数控机床
沙丘猫搜索算法
机器学习
随机配置网络
螺旋曲面
工艺参数
正交实验
加工能耗
CNC machine tool
sand cat swarm optimization
machine learning
stochastic configuration networks
helical surface
process parameter
orthogonal test
machining energy consumption
