NIPS 2018 Expo Demo
Dec. 8, 2019
Real-time Defect Detection with SAS Deep Learning and Event Stream Processing
Sponsor: SAS Institute Inc.
Manufacturing processes and product quality inspections require an ability to easily and quickly identify defective items. With the significant advancement in image recognition capabilities with powerful deep learning models, automated detection of defects is now more feasible and accurate than ever before. This demonstration shows how deep learning capabilities from SAS® can be integrated into the SAS® Event Stream Processing framework to score images taken directly off the “assembly line.” An assembly line is simulated using an electronic turntable system with bottles containing different types of liquids and with different “defects” related to caps and labels. A Raspberry Pi with a sensor triggers a camera to take a photo and send it to a laptop via a REST call. A CNN model is trained on a large set of images of these bottles in different states using the SAS DLPy Python module, with augmented images added. This model is then “deployed” to this system such that new bottles can be scored in terms of being defective or not.�