In the realm of industrial automation, acquiring extensive real-world data for training deep learning models presents a significant challenge, hindering the rapid development and deployment of advanced computer vision systems. This paper introduces a novel approach to address this limitation by leveraging the potential of synthetic data for training purposes. Focusing on a case study of a pick-and-place system, we detail the methodology for generating high-fidelity synthetic datasets from 3D models of tools and their subsequent use in training convolutional neural networks. Our research demonstrates that models trained on synthetic data not only fill the gap caused by the scarcity of real-world datasets but also significantly enhance the accuracy and efficiency of industrial automation systems. The results highlight the viability of synthetic datasets as a pivotal resource in the early stages of model training, offering a cost-effective and scalable alternative to traditional data acquisition methods.