A texture based descriptors used for real time tool condition monitoring

Tool Condition Monitoring Systems (TCMs), in the task of tool wear recognition, are highly dependent on the choice of descriptors. It is very important to extract the appropriate marking that contains information on the tool wear condition from the sensor signal. We consider the portions of the Short-Term Discrete Furrier Transform (STDFT) spectrum obtained by speaking from a vibration signal sensor as a 2D textured image. This is achieved by setting the STDFT timeline as the first dimension and the frequency scales as the second dimension of the resulting textured image. We divide the resulting textured image into special particular patches of texture, focusing on a portion of the frequency range of interest. Applying the set filter bank, 2D textons are allocated for each predefined frequency band. From 2D textons, for each frequency band, we extract probability density function (PDF) data in the form of lower order statistical moments. The applied method gives robust descriptors of the Tool wear state.