Extensive research in the field of artificial intelligence (AI) over the years has enabled advancements in versatile industry applications, including automated visual inspection, which represents an essential aspect of modern automation and manufacturing processes. This study demonstrates the use of deep learning for visual inspection and damage detection of ball bearings. We applied two deep learning techniques for detecting defects in bearings: classification based on Convolutional Neural Networks (CNN) and defect localization based on the You Only Look Once (YOLO) algorithm. We used the bearing dataset for defect localization obtained from Roboflow to train and evaluate three CNN models, including ResNet50, for classification. While these models achieved high accuracy, with ResNet50 reaching 99%, they lacked the ability to localize defects. In contrast, we trained YOLOv5 and YOLOv8 models for real-time defect detection and localization, with YOLOv5s providing the best test mAP@0.5 score of 86.1%, with efficient computation. Our study demonstrated that while CNN models had performed excellently in classification tasks, YOLO models provided the best balance between accurate detection and real-time localization, making them suitable for automated visual inspection in modern computer-integrated manufacturing (CIM) systems.