Enhancing Machine Vision Training with Digital Twins: A Toolchain for Optimized Image Categorization Using Synthetic Trainingsets

In dynamic production environments, frequent changes in products and setups demand efficient machine vision systems to verify part confgura-tions accurately. This study introduces a digital twin-enabled toolchain designed to generate synthetic images for machine vision training, thereby reducing dependence on large, real-world image datasets. By employing pretrained deep learning models and an extractor-based classification method, the toolchain significantly minimizes the required number of training images without compromising accuracy. An industrial case study reveals the effectiveness of this approach, achieving reliable performance with fewer images and reducing training time, ofering a cost-effective solution for adaptable, resilient manufacturing systems.