Manufacturing processes can be optimized by enabling human-robot collaboration. A relevant goal in this area is to create a collaborative solution in which robots can provide assisting actions to humans, thereby, reducing menial labor as well as increasing productivity. The solution is based on implementing efficient hand-over of mechanical tools from robots to humans. Hand-over tasks are inevitable in human-robot collaborative manufacturing contexts. These tasks need three-step mechanism: object identification, object grasping, and the actual hand-over. This paper presents an approach for robots to find tools for human partners in human-robot collaboration via deep learning. This is achieved using the object detection system YOLOv3 for identification of commonly used mechanical tools. By training on a custom dataset of 800 images of mechanical tools created for the study, the tool recognition is implemented in real-world human-robot hand-over tasks. Experimental results show that the proposed approach achieves a high accuracy for identification of tools in real-world human-robot collaboration. Future work of this study is also discussed.