Towards automatic generation of image recognition models for industrial robot arms

As the world moves towards mass customization, there is a need for a manufacturing system that can quickly adapt to market changes. Reconfigurable manufacturing systems (RMS) have been proposed as a solution. RMS is designed to be modular with a high degree of flexibility. However, such a structure creates a lot of complexity. For instance, if the modules are moved or changed, the robot arms in the system must be re-programmed. Adding 3D cameras and image recognition to the robot arms can solve some of these problems. Nevertheless, creating image recognition models is time-consuming work, requires human labor, and can increase the cost of manufacturing. To manufacture a large variety of products, there is a need to create image recognition models for each product. One method to automate the generation of image recognition models can be to use synthetic data. Synthetic data can be used to generate a large amount of labeled data, which can be used to train image recognition models.In this paper, we propose a method for training image recognition models using synthetic data, which can further automate robots in RMS. Specifically, the system utilizes a 3D model of a part to generate images, which are then processed by a cycle generative adversarial network (GAN) to enhance their realism. These images are subsequently auto-labeled and employed to train an image recognition model compatible with an industrial robot arm.