This study proposes a transfer learning (TL) pipeline for detecting surface defects in ball screw drives, which are vital in industries like robotics and aerospace. The pipeline uses pre-trained CNN models—InceptionV3, VGG16 and VGG19—to extract features from defect images, followed by classification with SVM (Support Vector Machine) and kNN (k-Nearest Neighbors). The InceptionV3 + SVM combination excels in accuracy, recall, precision, and F1 score, highlighting its effectiveness in defect detection. The study underscores the importance of selecting appropriate CNN architectures and classifiers for specific defect detection tasks. The dataset from the Karlsruhe Institute of Technology, consisting of 2000 images, is used to evaluate the TL pipeline's performance. The findings suggest that the InceptionV3 + SVM model offers a reliable method for identifying ball screw drive defects, with potential for further optimization through expanded datasets and hyperparameter tuning.