Applications of fuzzy K-NN in weld recognition and tool failure monitoring

Abstract

Two fuzzy K-NN (K-nearest neighbor) based procedures are developed for identifying welds from digitized radiographic images and for determining PCBN (polycrystalline cubic boron nitride) tool failure in face milling operations. Both procedures comprise two major components: feature extraction and fuzzy K-NN based pattern classification. For the weld identification application, the weld image is processed line-by-line and three features are extracted for each object in each line image. These features are: the width, the mean square error (MSE) between the object and its Gaussian, and the peak intensity. For the tool failure application, two features: /spl Delta/RMS and peak/count ratio, are derived from AE signals generated by the cutting operation. The use of the fuzzy K-NN classifier and the classification results are discussed. The results of this study indicate that the fuzzy K-NN based procedures produce a high successful rate of recognition for both applications.