基于AR模型的铣刀磨损诊断

传统的铣刀磨损故障诊断大多采用小波分析结合神经网络的方法，该方法的缺点是算法复杂，计算量大，很难实现铣刀磨损的在线识别并对其进行反馈控制。本文引入自回归（AR）模型来表征刀具切削过程的正常工作状态，用Levinson-Durbin递归算法求解Yule—Waker方程获得AR模型的系数。将建立的AR模型作为线性滤波器处理其它各种状态铣刀振动信号，获得预测误差信号，之后对预测误差信号进行各种统计特征分析。试验结果表明，预测误差信号的方差是有效的与刀具磨损相关的指标，可以用来在线识别加工过程铣刀磨损状态。

Diagnosis of Milling Cutter Condition Based on AR Model

Traditional failure diagnosis of milling cutter wear is mostly carried out by wavelet analysis combined with artificial neural networks （ANN）. However, this method has some shortcomings, such as complex arithmetic and a large amount of computation, and it is difficult to realize the on-line identification of milling cutter condition and carry out the feedback control. In this paper, the normal work condition in cutting process was described by auto-regressive （AR） model. The AR model coefficient was obtained by Levinson-Durbin recursion arithmetic. The AR model was used as a linear filter to process the vibration signals coming from the milling cutters under different wear condition, and the predicted error signals were obtained. Finally, statistical characteristics of the predicted error signals were analyzed. The results show that the variance of the predicted error signals is an effective characteristic index correlated with the cutting-tool wear, and it can be used to identify the milling cutter on-line condition in machining process.