Machine Vision AI, Self-learning Process Disturbance Monitoring Device

In the paper industry, there are numerous places where it would be beneficial to monitor and measure certain quantities, like issues on product quality, dirtiness on process equipment and process disturbances. Traditionally this has been done by operators, but continuous awareness is very tiring and practically impossible. Some of these locations are also dangerous to be in, the process is too fast to follow, or important events may be very random and rare. Applications have been developed based on machine vision, where computers analyze video feed from cameras and alarm users to do corrective action when certain quantities exceed pre-set limits. These applications are typically very tightly bound to one position, sensitive to external disturbances and very time consuming and costly to develop, so their usage is quite limited. During the past ten years new techniques have evolved, which allow self-learning from video feed and raising an alarm if there is something which does not fit to learned pattern. If the process or products changes, application can adapt to the changed situation. This paper presents research on detecting defects in products on corrugator machine using deep learning technology. The presented results and the experiments show potential of the methods in paper industry.