KICK-PV: AI-based characterization and classification of PV systems for predictive maintenance

Project background

So far only possible to a limited extent to detect faults in PV systems via remote maintenance. In most cases, faults are only detected relatively late and the extent can only be roughly estimated.

Project objective

The project AI methods are to be used to significantly increase the precision of remote maintenance of photovoltaic systems in order to identify fault patterns at an early stage and categorize them according to severity and cause at an early stage.

Operators of PV systems are thus protected from financial damage in good time and the calculability of the systems is improved. In addition, possible fault patterns are to be detected with a laboratory developed as part of the project in the course of on-site analyses. precisely classified and quantified in detail down to module level if possible.

Project procedure

1. Data analysis and filtering of existing data streams at Smartblue AG
2. Improvement of predictive maintenance maintenance with the help of AI methods. Goal: Detecting faults as early as possible and classification of faults according to severity and cause.
3. Development of a mobile on-site measurement laboratory to detect faulty modules within larger module strings (serial module interconnection) and to quantify individual module losses.

Innovation

• Combining physical models for performance prediction with AI methods.
• AI methods should be embedded in explainable models and thus errors can be errors can be classified according to severity and cause.
• Conventional AI models also provide performance predictions, however (black box) it is usually impossible to quantitatively classify losses according to cause.