Presed – Predictive Sensor Data mining

Initial situation:

• Some steelworks experiences problems with product deficiencies like slivers and cracks.
• Several pre-studies indicated a relationship between the time dependence of a set of measured values and the occurrence frequency of the defect.
• Methods for predicting, detecting and reducing the defect early in the process are desired.

Working topics:

• Highly resolved time-series process data is analysed with respect to the root causes for these defects.
• Influential key factors are derived from the time-series data to quantify the probability of the defects.
• An event-detection for time-series data will be developed.
• The know-how about the interdependencies is conserved in an ontological knowledge base.

Results:

• A prediction system for the defect probability is developed to assist the plant personnel.
• Novel results on defect appearance and root causes are expected.
• Reduction of slivers and cracks by operating the processes at optimum working points.