New sensor technology should significantly increase the efficiency of large solar power plants. The technology detects failures early and corrects them. In this way, the system reduces downtime, which increases solar farm yields.
The development is part of the ZeroDefect4PV project by Romania's BEIA Consult International, Turkey's INELSO Innovative Electrical Solutions and Germany's Fraunhofer IFF. Large solar power plants often consist of tens of thousands of modules and components. Faults in individual modules can significantly reduce the overall efficiency of strings connected together.
However, the exact location of faults in individual panels or clusters is difficult to detect with current technology. Researchers are currently testing an integrated sensor prototype, developed by INELSO Innovative Electrical Solutions, that should make this possible. The sensor system works with an intelligent IoT communication architecture, providing module-level insight.
Detecting inefficiencies
The system works with a network of sensors placed in the solar farm. These 'Data Collection Units' communicate via a mesh network, where the individual units are connected to each other. Thus, they transmit data to a server, which uses AI, among other things, to analyse collected data. For example, the system can be used to detect anomalies, contamination and defects early.
Within the project, Fraunhofer IFF focuses on the intercommunication of the sensor prototypes and extracting insights from the collected measurement data. It focuses on:
- Data completion: Missing data in datasets can hinder follow-up processes. Using AI, missing data is filled with minimal deviations from actual values. In the process, the models continuously learn based on new data.
- Anomaly detection: Models use mathematical characteristics to describe the supplied data. By defining confidence intervals, anomalies are identified that indicate malfunctions or plant failures.
- Predictive maintenance: Models detect anomalies, in some cases before they occur. This allows faults to be detected and resolved faster.
In addition, Fraunhofer IFF is investigating how information collected with the sensor network can be used for solar farm management functions.
'More transparency, better forecast quality and reliability'
Hannes Peter Wasser, researcher at Fraunhofer IFF: "Every kilowatt hour from renewable energy that is not imported increases the need for fossil balance energy and works against climate mitigation goals. The growing system relevance of photovoltaics requires more transparency, better forecast quality and reliability. We will achieve this in the future with our solution, which combines a high-resolution module-level sensor system with AI diagnostics, prediction and anomaly detection modes and a modular platform for collecting, synchronising, pre-processing and storing all data."
Dr Christoph Wenge, researcher at Fraunhofer IFF, adds: "A wide range of faults can occur in a solar panel string, not only in the modules themselves, but also in the bypass diodes, wiring or mounting systems. Unlike measurements at the inverter, our system classifies faults. It detects where they occur. AI models, pre-trained with different faults, analyse patterns, identify deviations from normal performance, recognise anomalies and their impact, e.g. whether string A delivers less power than string B. Implemented support functions, displayed on monitors in the control room, provide staff with recommended actions, such as cleaning or replacing a module."
Pilot system
Tests are currently being carried out on the pilot system at Fraunhofer IFF. Researchers are testing whether AI models can recognise types of faults based on small changes in current and voltage curves. Sensors have previously been tested in the lab for measurement accuracy, stability and communication reliability. Other tests include targeted shading of modules and interpretation of thermal images.
Tests are also taking place at INELSO Innovative Electrical Solutions in Turkey, focusing on hardware validation in a solar field. BEIA Consult International is testing Fraunhofer IFF's AI models in Romania with SolarEdge inverter data on energy consumption and production.
Wenge: "With the pilot installations, we are validating our system under realistic conditions so that we can iteratively optimise hardware, communication and data models and assess scalability for large solar farms."
