摘要
刀具的磨损状态直接影响产品加工质量、成本和效率,对刀具磨损量的实时监测识别具有重要意义。针对刀具磨损状态先验样本少和常规神经网络识别模型收敛速度慢、易陷入局部极小值等问题,提出了基于最小二乘支持向量机(LS-SVM)的刀具磨损识别方法,并针对支持向量机的惩罚因子和核参数对模型识别精度影响较大的问题,提出一种根据个体适应度来调整惯性权重的自适应粒子群算法进行自动参数寻优。以车削加工为研究对象,采集加工过程中的切削力信号,应用小波包分析技术提取反映刀具磨损状态的特征信息作为识别模型的输入,然后利用训练好的自适应粒子群算法优化后的LS-SVM识别模型进行刀具磨损量识别。实验结果表明,该自适应粒子群优化算法比标准粒子群优化算法参数寻优能力更强;粒子群优化LS-SVM模型能高效地实现刀具磨损量识别,与BP神经网络相比具有更高的精度,且所需样本数较少,训练速度更快。
Tool wear state directly affects the product quality, pcost and efficiency. The real-time condition motioring system for tool wear would be significant. The prior samples for monitoring model were limited, and the conventional neural networks recognition model had some drawbacks such running into local minimum value easily, slow convergence rate and so on. In view of these situations, it proposed a tool wear monitoring method based on least squares support vector machine (LS-SVM). Meanwhile,it proposed the adaptive particle swarm optimization (APSO) algorithm to search optimum value of the kernel func- tion parameter and error penalty factor which affected the precision of the LS-SVM significantly. The cutting force signals were measured as monitoring signals. Features extracted by wavelet package decomposition as model inputs. Tool wear degree could be got by the trained APS0-LS-SYM model. The experiment result shows APSO algorithm can find the better parameters of LS- SVM model than standard PSO algorithm;APS0-LS-SVM carry out an accurate tool wear assessment, unlike BP neural net- works, the algorithms have higher accuracy, faster learning ability and needs less training samples.
出处
《计算机应用研究》
CSCD
北大核心
2014年第4期1094-1097,1101,共5页
Application Research of Computers
基金
国家科技重大专项基金资助项目(2010ZX04015-011)
中央高校基本科研业务费专项基金资助项目(SWJTU12CX039)
关键词
刀具状态监测
小波包分析
粒子群优化
最小二乘支持向量机
tool condition monitoring
wavelet packet analysis
particle swarm optimization
least squares support vector machine