Vision-based system for automated image dataset labelling and dimension measurements on shop floor

Vision-based systems augmented with deep learning-based Convolutional Neural Networks (CNNs) can effectively capture process or product deviations and are vital for achieving smartness in manufacturing. Acquiring good-quality component images on the shop floor is challenging yet mandatory for labeled dataset generation while developing CNN models. The present work develops a vision-based system utilizing hardware and software to capture good-quality images of manufactured components. The system can perform onboard image pre-processing efficiently and generate labeled image datasets. The experiments are performed to capture component images with different lighting conditions. It is shown that training of CNN-based image classification algorithm using images acquired by the developed system achieves better prediction accuracy. The developed system can also perform dimension measurement tasks employing classical image processing modules. The study showed that the system could be effectively implemented for image-based dimensional metrology and labeled dataset generation tasks offering ease of operation, portability, robustness, and versatility.