The growing complexity of production plants and therefore of automation systems generates a need for an automatic detection of non-normal situations such as errors, degradation effects, or suboptimal energy consumptions. If such situations could be detected early, downtimes, ill-scheduled repairs, or high costs could be avoided. In this project, self-learning and therefore adaptable anomaly detection and diagnosis algorithms will be developed and will be integrated into existing automation systems. These approaches (i) are able to learn errors or suboptimal situations, (ii) can automatically identify error causes and therefore repair strategies, and (iii) are based on only minimal manual engineering efforts. The results will be applied to plants in the field of process engineering and in the field of manufacturing engineering.

Proposal Outline:

Solution Approach: 1) In a first step, the data acquisition problem is tackled. For this, data from all parts of the plant (sensors, actors, MES-systems, ERP-systems, energy consumptions) must be gathers and synchronized. As a result, plant operators are able to visualize a synchronized overall status of their plant. 2) In a second step, these data is recorded, analyzed and abstracted. This is done by means of statistics, machine learning, and data mining. At the end of this step, a model of the normal and of the faulty plant behavior over a period of time is identified. This model can already be used to predict the plant behavior and to identify optimization potentials. 3) Finally, by comparing the real plant behavior to the behavior predicted by the learned models, non-normal or suboptimal situations can be identified automatically during the plant operation. Furthermore, error causes can also be identified.

For further information, please contact: jaime.duran@juntadeandalucia.es