Comparison of Recognition Techniques to Classify Wear Particle Texture

Wear particle analysis, which identifies failure modes caused by the wear of various machine components, is an essential technique for monitoring machinery conditions. This analysis plays a vital role in predictive maintenance by revealing component degradation in machinery. This study proposes an automated framework to classify four standard wear particle textures—rough, striated, pitted, and fatigued—using artificial neural networks (ANNs) combined with advanced image processing techniques. Images acquired via Charged-Coupled Device (CCD) microscopy were preprocessed using sharpening, histogram stretching, and four edge detection algorithms: Sobel, Laplacian, Boie–Cox, and Canny. The Laplacian and Canny methods yielded the highest classification accuracies of 97.9% and 98.9%, respectively. By minimizing human subjectivity, this automated approach enhances diagnostic consistency and represents a scalable solution for industrial condition monitoring.