Use case AI: Early detection of errors and abnormalities in the production

X-Pact^® MES 4.0 utilizes innovative methods of artificial intelligence. These include machine learning and big data analyses. Different methods are used in several MES system modules and provide, for example:

• Early detection of abnormalities in the production
• Due date prediction,
• Autosequencing,
• Throughput prediction

The following video explains the early detection of errors and abnormalities during production using the example of a CSP^® plant. With this process, steel melts are cast into slabs by the caster, then rolled straight afterwards to produce strip, which is wound into coils.

Artificial intelligence is capable of detecting unknown correlations between different input parameters and thereby identifying disturbance factors at an early stage. This allows action to be taken and economic damage to be limited.

In the present case it was noticed that the strip temperatures on the coiler at times suddenly differed considerably from the setpoint value. As such temperature deviations adversely affect the material properties of the coils, the cooling water volume was automatically adapted for the next-in-line strip, causing the difference between the actual and desired temperature to be reduced even further. Nevertheless, the temperature deviations experienced with the previously produced strips meant that these strips had to be downgraded.

Using the X-Pact^® Performance Enrichment Analysis – a method of artificial intelligence developed by SMS group and Jacobs University – the unexpected correlation between a faulty work roll in a mill stand and the temperature deviations during cooling can be demonstrated not only precisely, but also much more clearly and effectively than with a standard analysis. Once the Performance Enrichment Analysis module is activated, a whole range of possible errors can be simultaneously monitored.

How exactly does the X-Pact^® Performance Enrichment Analysis work?

The Performance Enrichment Analysis from SMS group continuously monitors a variety of input parameters, in order to detect any abnormalities during production. This is where techniques such as the cluster analysis, which groups similar data points into clusters, are used. In the specific use case in question, one data point represents a strip with its related process parameters and properties.

Various different parameters are examined for each coil produced, several cluster analyses are performed, and the temperature deviations in the clusters are determined.

The significance of the number of temperature deviations per cluster is then analyzed. The z value, as it is known, is calculated for this purpose. A high absolute z value indicates that there is a significant correlation between the respective temperature deviation and the characteristics of the cluster concerned (e.g. “thin strip”, “high furnace outlet temperature”, etc.), and that, in all probability, this cannot be put down to coincidence.

With regard to the input parameters (casting speed, furnace outlet temperature, etc.), no significant correlation with the temperature deviation could be found in the present case in question. By contrast, the analysis of the roll IDs showed that the deviation between the setpoint and actual temperature was high with a particularly large number of coils, in cases where the work roll with ID 44 was used or was replaced during the last roll change.

The evaluation of the data using the X-Pact^® Performance Enrichment Analysis is continuously performed during plant operation, thus allowing it to detect deviations during production quickly and send reports on the abnormalities. At the same time, a variety of other target variables can be monitored instead of the coiler temperature.

X-Pact^® Business Intelligence automatically generates reports which are used by the personnel in charge to examine the errors in greater detail, where necessary, and ultimately to eliminate them.