Online Gear Wear Particle Detection and Categorization Using a Convolutional Neural Network Algorithm Integrated with Cascade Classifier

Most gears fail because of wear caused by rubbing, metal to metal contact, contamination, or breakdown of lubrication. Because of this, figuring out how to find and sort wear debris particles is an important area of research for both predictive and proactive maintenance. By putting these wear particles into different categories like spherical, cutting, fatigue, sliding and rubbing, it would be possible to identify the wear modes present in the gearbox and predict the nature of failure and condition of the system. The present research aims to automate the detection and classification process using the Convolutional Neural Network (CNN) integrated with Cascade classifier. CNN automatically extracts different suitable features from images by applying multiple filters on it and also reduces the complexity of image processing whereas the Cascade classifier is used to detect the particles by differentiating between positive and negative images by applying the Haar-like features into it. The objective of the research work is to provide a most efficient and accurate detection and classification of wear debris particles using a trained cascade classifier integrated with a customized lightweight CNN model named as Wear Particle Classifier Net (WPCnet).