基于深度卷积神经网络的刀具磨损量自动提取方法

2019-12-16

作者：李德华1，刘献礼2 单位：1哈尔滨理工大学测控技术与通信工程学院；2哈尔滨理工大学机械动力工程学院

摘要：刀具状态检测可以有效降低加工过程中刀具的不确定性，提高数控加工质量和效率，降低加工成本。在小批量制造模式下的复杂零件制造过程中，零件的几何形状和加工参数不断变化，刀具所受外力也在不断改变，进而导致刀具磨损速率持续变化。传统的固定切削时间更换刀具的方法只能采取更加保守的切削时间更换刀具，给加工过程增加了很多的不确定性，并造成严重的刀具浪费。本文针对以上问题提出了一种刀具磨损在线测量方法，通过电子显微镜在线拍摄刀具照片，经小波滤波降噪处理后的图片由卷积神经网络进行处理，并自动计算出刀具磨损量。该方法可以有效地提取出刀具磨损量，测量误差不超过0.02mm。

关键词：刀具磨损；深度学习；卷积神经网络；小波滤波；自动计算

中图分类号：TG713；TG529文献标志码：ADOI：10.3969/j.issn.1000-7008.2019.12.020

Automatic Tool Wear Extraction Method Based on

Deep Convolutional Neural Network

Li Dehua，Liu Xianli

Abstract:Tool state detection can effectively reduce the uncertainty of the tool during machining，improve the quality and efficiency of CNC machining and reduce the processing cost.In the manufacturing process，the complex parts in the small batch manufacturing mode，the geometry and machining parameters of the parts are constantly changing，and the continuous change of the external force of the tool causes the tool wear rate to continuously change.The traditional method of changing the tool for fixed cutting time can only replace the tool with a more conservative cutting time，this process adds a lot of uncertainties and causes serious tool waste.An online measurement method of tool wear is proposed for the above situation.In this method，the tool photo is taken online by an electron microscope，and the noise is reduced by wavelet filter，and the processed image is processed by the convolutional neural network to automatically calculate tool wear.This method can effectively extract the tool wear amount，measurement error does not exceed 0.02mm.

Keywords:tool wear；deep learning；convolutional neural network；wavelet filter；automatic calculation