Real-time image processing, Industrial signal processing, Pattern recognition

MaDiSec: Machine Diagnosis for Security Printing Machines

01.07.2010 bis 30.06.2012

Current state-of-the-art inspection systems are in most cases only based on one physical sensor value. Pressure, force, temperature, optical properties or others are acquired and individually evaluated. Also if such sensor information is processed perfectly, the quality of inspection, particularly the early recognition of consecutive errors, is often not sufficient. The fusion of several information sources may allow precise statements regarding the machine condition.

The target of this project is the detection of errors on security printing machines, in particular on steel engraving machines. The main focal point is the early recognition of consecutive errors in order to avoid printing errors by combining measuring data with expert knowledge. By assessing the expert knowledge, it is important to pay attention to the fact that in addition to explicit information which could be retrieved directly, also implicit knowledge, consisting of routines, experience and subconsciously stored knowledge could be acquired.

The choice of feasible information sources plays an important role within the design of inspection systems. In doing so it is important that measuring variables, positions and directions are chosen advisedly, so that effects in the machine become visible in the measured data. The influences of single variables as well as the interaction of several information sources are analyzed using statistical methods. The redundancy within the sensor data is used to check the reliability of the information and to point out defect information sources.

The classification of the manufactured product gives a feedback to the inspection system and provides extension of the learned model. In the case of the printing machine the task is performed by the optical inspection unit.

This project is promoted by:
Stakeholders / Contacts: Karl Voth, M. Sc., Prof. Dr.-Ing. Volker Lohweg
Investigations on Possibilistic Multi-source Data Fusion with Sensor Reliability Monitoring
Stefan Glock, M. Sc.