Detection of Defective Products in Injection Molding Process using YOLO-NAS

Injection molding is an important manufacturing process with applications spanning various industries. Modern industry demands quality, efficiency, and environmental sustainability. The advanced injection molding inspection systems (IMIS) provide these requirements by real-time monitoring systems executed based on Industry 4.0 technologies such as IoT, machine vision, and artificial intelligence. This paper introduces an IMIS using YOLO-NAS models as one of the state-of-the-art deep neural networks for real-time object detection, to recognize both complete and defective injected products. Upon defective products detection, the system automatically stops the injection process, ensuring operational safety and product quality. Our evaluation of implemented models based on YOLO-NAS variants (i.e. small, medium, large) for injection molding inspection reveals minor performance differences. The results declare that YOLO-NAS models have a reliable performance on defective products detection in an injection molding machine.