Artificial Intelligence for Monitoring the Status of PV Systems

Photovoltaic systems provide a substantial contribution to sustainable power supply in Germany. However, not all modules and components in all systems function flawlessly at all times. Potential weaknesses or defects can lead to yield losses. If they are not identified in a timely manner, this can result in significant economic damage for the operator. Currently, however, it is not always possible to tailor fault detection to each individual module, taking into account the specific characteristics of the material selection, manufacturing processes, installation, or location. There is an urgent need for monitoring solutions that enable damage to be detected at an early stage and corrected without major economic loss.

This is where the joint project “Mon-KI” undertaken by GETEC green energy GmbH Magdeburg, which develops renewable energy supply solutions, and Fraunhofer CSP came in. In the two-year project, the use of AI methods enabled better prediction of yields and maintenance work on photovoltaic modules.

The scientists at Fraunhofer CSP used computer-assisted methods to validate actual and target data. Based on field inspections, historical data, and laboratory tests, defect patterns in monitoring data were recorded for training machine learning models. This enabled the researchers to automatically detect defects that lead to degradation and yield loss in PV modules. The research team considered potential-induced degradation (PID), light- and temperature-induced degradation (LID and LeTiD), bypass diode faults, AC fuse faults, cell breaks, and shading and partial shading caused by dirt, vegetation, near and far shadows, and snow as defects.

The characterization of the target state was used to create ideal images of photovoltaic systems with regard to individual components such as modules, cables, inverters, electrical wiring, topography, and weather conditions in order to forecast performance and energy yield. A distinction was made between new systems, which are expected to be defect-free at the start of operation, and existing systems that are already in operation and may be subject to defects.

The actual data was provided by a monitoring box prototype developed as part of the project, which used various sensors to record environmental parameters, thereby enabling different PV systems to be compared, identified, and quantified. Dr. Matthias Ebert, group manager for “Module and System Reliability” at Fraunhofer CSP, commented on the start of the project: “We are confident that the project will solve fundamental problems in current photovoltaic monitoring and reduce the currently very time-consuming manual analysis of defect data. By using AI methods, we are creating reliable operation of PV systems by identifying defects in a timely manner, thereby optimizing and reducing maintenance costs. The results can be used to develop new business models.”