The work on the control tool was focused on the data mining module. Within this module an algorithm was developed which allows to automatically detect faults in solar thermal systems based on artificial intelligence methods. The algorithm uses a so-called Random-Forest-Regression method to identify the non-linear relationships between different sensors and models their behaviour. Faults can then be detected by comparing the predicted values with new measurement data and alarms can be raised if the difference exceeds a confidence threshold. The algorithm offers the following advantages:

  • Minimal configuration effort – since it automatically identifies highly correlated features
  • High fault-coverage – as indicated by preliminary results on a test data set

The ability to detect faults is shown in the following figure for the temperature of a solar thermal system. It can be seen that the algorithm automatically and very early detects high temperatures in the dataset and raises an alarm (red points) as soon as the difference between prediction and measurements is too high.

Figure 1: Preliminary results of the fault detection algorithm for a solar collector system


In the following the algorithm should be further improved, tested and implemented.