This works aims at detecting ball screw driver deffects as a Computer Vision task based on  Deep Learning single-shot YOLO models. Computational experiments using a realistic dataset were performed and their results altogether with statistical tests enlisted YOLOv5 Nano as best suited for this purpose, with an average mAP@0.5 of 93.27% and with improvements of 97.84% on speed and on  95.68% less parameters than a related work counterpart. The proposed solution is recommended for embedded devices and favors equipment monitoring for Industry 4.0, also suggesting improvements in state of art in terms of efficiency of deffects detection.