PV fault detection/monitoring system using machine-learning algorithms. A learning algorithm is trained to estimate power production based on meteorological data (irradiance, temperature, wind speed, etc.). This predic-tion is then compared to the actual power to detect various faults. Current research is focused on distinguishing different fault signatures. (Figure from Riley and Johnson, 2012)

PV fault detection/monitoring system using machine-learning algorithms. A learning algorithm is trained to estimate power production based on meteorological data (irradiance, temperature, wind speed, etc.). This predic-tion is then compared to the actual power to detect various faults. Current research is focused on distinguishing different fault signatures. (Figure from Riley and Johnson, 2012)

Sandia researcher Joshua Stein (in Sandia’s Photovoltaic & Distributed Systems Integration Dept.) highlighted novel PV array monitoring strategies at a PV Performance Analysis and Module Reliability workshop held at the European PV Solar Energy Conference and Exhibition in Amsterdam, Netherlands. The talk was coauthored by Mike Green from MG Lightning Electrical Engineering in Israel.

Under the International Energy Agency Photovoltaic Power Systems (PVPS) program, Task 13, new methods to monitor PV systems are actively being developed

  • machine-learning algorithms,
  • new DC-monitoring applications,
  • fault detection and location, and
  • prognostics & health management (PHM) techniques.
Predicting faults using machine-learning algorithms. Under the assumption that all inverter parameters behave the same under normal conditions, when the normal behavior changes, a fault is on the way. (This concept is under development in PVPS Task 13, Activity 2.2, led by Mike Green.)

Predicting faults using machine-learning algorithms. Under the assumption that all inverter parameters behave the same under normal conditions, when the normal behavior changes, a fault is on the way. (This concept is under development in PVPS Task 13, Activity 2.2, led by Mike Green.)

Our research has found that mere data collection is not enough; rapid, automated, and integrated analysis methods are required to derive intelligence on system performance and health. Future monitoring systems will be seamlessly integrated into operations and maintenance (O&M) activities.

Our ultimate goal is PHM that is directly integrated into the PV system and possibly even individual components.

  • Low-cost sensors with embedded and central data processing to monitor system performance and predict present and future problems (e.g., engine check/service light).
  • Little to no energy is lost from unplanned outages.
  • O&M activities are optimized to maximize availability.
  • Monitoring system costs are low enough to justify investment.

The workshop was organized by the IEA PVPS program, Task 13 to discuss the work of the task experts over the last four years and first results from the second phase of the Task 13 work program.

An article summarizing Stein and Green’s talk is forthcoming in PV-Tech magazine.