摘要
在线刀具磨损量估算及其未来发展趋势预测对于指导现实生产有着十分重要的意义。提出基于L-M优化算法BP神经网络的刀具磨损量在线预测方法。对声发射信号进行小波包分解,得到32个不同频带内的信号,用于构造初始特征向量矩阵;对初始特征向量矩阵进行奇异值分解,计算奇异谱,将奇异谱做为刀具磨损的特征向量,利用神经网络在线预测刀具磨损量。试验结果表明:预测结果能准确地跟踪实际的刀具磨损曲线,并且L-M优化算法比其他改进算法迭代次数少,收敛速度快,精确度高。
Online tool wear estimation and future trends prediction have great significance in guiding the practical production. Online tool wear prediction method based on improved L-M optimizational algorithm BP neural network was proposed, By wavelet packet decomposition, the collected acoustic emission signals were decomposed into 32 signals with different frequency bands, which were used to construct the initial feature vector matrix. Then using initial feature vector matrix, the singular spectrum was calculated by sin- gular value decomposition. Feature vector constituted by the singular spectrum was used to realize tool flank wear VB value forecasting using BP neural network with L-M algorithm. The experimental results show that prediction result can accurately track the actual tool wear curve. The BP neural network based on the improved L-M optimizational algorithm has merits of fewer iterations, fast conver- gence and high accuracy.
出处
《机床与液压》
北大核心
2012年第15期22-26,共5页
Machine Tool & Hydraulics
基金
东北电力大学博士科研启动基金资助项目(BSJXM-201115)
关键词
刀具磨损量预测
L-M优化算法
BP神经网络
小波包分解
奇异值分解
Tool wear prediction
L-M optimizational algorithm
BP neural network
Wavelet packet decomposition
Singularvalue decomposition
