基于改进的径向基神经网络刀具磨损识别方法
|
摘要
采用传统的径向基神经网络(RBF)模型对刀具磨损量进行识别,没有体现铣削力特征值与刀具磨损量关联程度。文章通过特征优选获取最能反映刀具磨损状态的特征,并将优选后得到的特征与刀具磨损量的关联度作为对角矩阵的对角元,再将该对角矩阵引入RBF径向基函数中,从而使得在计算RBF输入向量与中心神经元径向距离时,不同特征的空间距离所占比重等于该特征对应的与刀具磨损状态的关联度,代替原RBF计算中不同特征的空间距离所占比重相等的情况。实验证明,与优化前相比,优化后的RBF神经网络模型具有更高的识别准确率和识别速度。
The traditional radial basis neural network(RBF) model was used to identify tool wear, but the correlation between milling force and tool wear was not shown. In this paper, the characteristics that best reflect tool wear are obtained by feature optimization. The values of degree of association between the features obtained after optimization and the amount of tool wear are used as the diagonal elements of the diagonal matrix. The diagonal matrix is introduced into RBF radial basis function, so that the proportion of the spatial distance of different features is equal to the correlation degree between the features and the tool wear state while calculating the radial distance between RBF input vectors and central neurons, instead of the situation that proportion of the spatial distance of different features equal in the calculation of original RBF. The result of experiment indicates that the optimized RBF neural network model has higher recognition accuracy and recognition speed than that before optimization.
The traditional radial basis neural network(RBF) model was used to identify tool wear, but the correlation between milling force and tool wear was not shown. In this paper, the characteristics that best reflect tool wear are obtained by feature optimization. The values of degree of association between the features obtained after optimization and the amount of tool wear are used as the diagonal elements of the diagonal matrix. The diagonal matrix is introduced into RBF radial basis function, so that the proportion of the spatial distance of different features is equal to the correlation degree between the features and the tool wear state while calculating the radial distance between RBF input vectors and central neurons, instead of the situation that proportion of the spatial distance of different features equal in the calculation of original RBF. The result of experiment indicates that the optimized RBF neural network model has higher recognition accuracy and recognition speed than that before optimization.
引文
[1] S Das,A B Chattopadhyay, A S R Murthy.Force Parametersfor On-line Tool Wear Estimation:A Neural Network Approach[J].Neural Networks,1996,9(9):1639-1645.
[2] Y K Koren,Tsu-Ren,GalipUlsoy,et al.Flank wear eastimation varying cutting condition[J].Joural of Dynamic Systems,Measurement and control,Transaction of the ASME,1991,113(2):300-307.
[3] A G Trustly J.A critical review of sensors for unmanned maching[J].Annals of CRIP,1983,32(2):563-572.
[4] 王姣,祁美玲. RBF云神经网络在数控机床刀具磨损状态识别中的应用[J]. 机床与液压,2011,39(15):146-150.
[5] 丁小峰,周月超. 基于RBF神经网络的刀具磨损量的估测[J].机床与液压,2005(10):59-62.
[6] 张强,刘志恒,代晓丹,等.基于振动和声发射信号的截齿磨损程度识别研究[J]. 机械设计与研究,2018,34(1):125-132.
[7] 蒋丽英,王蕾,席剑辉,等.基于RBF网络的刀具磨损状态预测技术研究[J].仪器仪表学报,2009,30(6):177-199.
[8] 刘文菊,郭景.RBF神经网络中心选取OLS算法的研究[J].天津工业大学学报,2002,21(2):71-73.
[9] 刘姝琴,廖晓昕.RBF神经网络的一种鲁棒学习算法[J].华中理工大学学报,2000,28(2):8-10.
[10] 袁曾任.人工神经网络及其应用[M].北京:清华大学出版社,2000.
[11] 黄华,李爱平.基于小波神经网络的切削刀具磨损识别,[J].农业机械学报,2008,39( 8):173-177.
[2] Y K Koren,Tsu-Ren,GalipUlsoy,et al.Flank wear eastimation varying cutting condition[J].Joural of Dynamic Systems,Measurement and control,Transaction of the ASME,1991,113(2):300-307.
[3] A G Trustly J.A critical review of sensors for unmanned maching[J].Annals of CRIP,1983,32(2):563-572.
[4] 王姣,祁美玲. RBF云神经网络在数控机床刀具磨损状态识别中的应用[J]. 机床与液压,2011,39(15):146-150.
[5] 丁小峰,周月超. 基于RBF神经网络的刀具磨损量的估测[J].机床与液压,2005(10):59-62.
[6] 张强,刘志恒,代晓丹,等.基于振动和声发射信号的截齿磨损程度识别研究[J]. 机械设计与研究,2018,34(1):125-132.
[7] 蒋丽英,王蕾,席剑辉,等.基于RBF网络的刀具磨损状态预测技术研究[J].仪器仪表学报,2009,30(6):177-199.
[8] 刘文菊,郭景.RBF神经网络中心选取OLS算法的研究[J].天津工业大学学报,2002,21(2):71-73.
[9] 刘姝琴,廖晓昕.RBF神经网络的一种鲁棒学习算法[J].华中理工大学学报,2000,28(2):8-10.
[10] 袁曾任.人工神经网络及其应用[M].北京:清华大学出版社,2000.
[11] 黄华,李爱平.基于小波神经网络的切削刀具磨损识别,[J].农业机械学报,2008,39( 8):173-177.