基于视觉/声发射多源特征分析与融合的刀具状态识别研究

It is very important for the trouble-free operation of the machine equipment and improving product quality by grasping the qualitative and quantitative of the tool condition detection and diagnosing tool wear, chipping and other damage fault. In this project, the visual features of the work piece machined surface texture images were analyzed by the tool in different wear or breakage state. At the same time, cutting acoustic emission signal feature extraction was also provided. The new theories and methods for tool condition monitoring were put forward by multi-feature extraction and classification fusion technology. The main research points can be concluded as following. The first, the study based on the AM-FM model of the work piece surface texture feature extraction was applied from the point of view of the signal processing, and then the high frequency information in the direction perpendicular to the work piece and the tool path were also extracted by wavelet packet transform. The second, modern signal processing methods such as the empirical mode decomposition, higher order spectral analysis, and wavelet analysis were introduced in the signal processing of tool wear or breakage. The joint characteristics vector was structured from signals in time domain, frequency domain, time-frequency two-domain feature extraction. For the last, the fusion feature extraction based on Kernel Principal Component Analysis was studied, and the tool wear or breakage state classification based on support vector machine was used. Finally, the tool wear or breakage state of qualitative and quantitative identification was realized by multi-feature fusion. From this project study, the tool wear condition monitoring system solution will be established for practical industrial application in the future.

定性、定量地掌握刀具状态，检测与诊断刀具磨损、崩刃等损伤故障，对于延长机床设备无故障运行，提高产品质量具有重要意义。本课题围绕刀具在不同磨损、破损状态下工件加工表面图像表现出来的视觉特征，以及切削声发射信号的特征提取，通过多特征提取与信息分类融合技术，为刀具状态监测提供新的理论与方法。具体内容包括：（1）从信号处理的角度，研究基于AM-FM模型的工件加工表面图像纹理特征提取方法，利用小波包变换，提取工件与刀具轨迹正交的方向上的高频信息。（2）将经验模态分解、高阶谱分析和小波分析等现代信号处理方法引入刀具磨损、破损信号的处理，分别从信号的时域、频域、时-频两域进行特征提取，构造联合特征向量。（3）采用基于核主元分析法的融合特征提取和基于支持向量机的刀具磨损、破损状态分类研究，实现多特征融合的刀具磨损、破损状态的定性、定量识别，为构建和开发通用的刀具监测系统解决方案奠定理论基础。

刀具是机械加工的直接执行者，随着加工过程的逐步进行，刀具出现磨损现象是不可避免的，其工作状态的好坏直接影响着切削精度、生产效率以及产品质量。因此，如何定性、定量地掌握刀具状态，检测与诊断刀具磨损、崩刃等损伤故障，对于延长机床设备无故障运行，提高产品质量具有重要意义。. 本课题针对刀具在不同磨损、破损状态下切削时声发射信号的特点，从传统的方法入手（特征提取-状态识别）入手，并引入大数据背景下的深度学习技术，为刀具状态监测提供新的理论与方法。具体内容包括：（1）从信号处理的角度，研究了基于互补总体平均经验模态分解(Complementary Ensemble Empirical Mode Decomposition, CEEMD)和小波包变换(Wavelet Package Transform, WPT)的特征提取方法，实现了对声发射信号特征分量的精确提取。（2）在信号特征提取的基础上，采用支持向量机，完成对刀具磨损、破损状态的分类研究。（3）针对声发射信号数据量大的特点，将堆叠自编码网络、卷积神经网络等深度学习方法引入到刀具磨损、破损信号的处理中，以刀具的时域、频域信号作为网络输入，通过深度网络自适应提取声发射信号特征，实现刀具磨损、破损状态的定性、定量识别，为构建和开发通用的刀具监测系统解决方案奠定理论基础。