Applying SSD to Real World Food Production Environments

To reduce the number of unplanned machine stops in packaging machines the root cause of the short but frequent interruptions has to be found. To assist operators in trouble shooting the current setting has to be automatically described to be able to link it with suitable knowledge bases. To start building the system the applicability of object detection using the deep learning architecture ""Single Shot Detector"" (SSD) is analyzed under real world conditions on a form-, fill- and sealing-machine. The objective is to have suitable coordinates of packaging goods and machine components to later calculate trajectories and doing anomaly detection on those trajectories to identify and distinguish different break downs.