Rolling Contact Bearing Fault Detection System Using Deep Learning

Recently, many new technologies are being used in the manufacturing industry. Quality control holds a significant position across all industries. Deep learning techniques aid manufacturers in evaluating product quality, thereby reducing the quantity of defective items that make their way into the market. The research mainly focused on detection of conditions of the bearing. The reliability and lifespan of rolling bearings are greatly influenced by wear, a significant tribological process. Through field examination of bearing failures resulting from wear, possible causes are identified, leading to necessary measures for reducing or eliminating wear. The process comprises collection of the images of surface of the bearing that includes possible defects identification with literature survey. Preprocessing of the collected images of both good and defective bearings was completed, and an image processing algorithm was developed using deep learning. The system was designed to analyze microscopic images and determine whether a bearing is in good condition or defective. Experimental results indicated the training dataset, cross-validation accuracy 89.91% and 91.89%, respectively.