Solar photovoltaic (PV) panels can last for 25 years or more, but before they hit the market they must undergo testing and meet stringent reliability standards. Now researchers at Case Western Reserve University are pairing advanced imaging techniques with statistical analyses to get an even deeper look into the damages—not visible to the eye—that lead to power loss over a panel’s lifetime.

To make sure the panels work as expected, scientists typically study them in various climates and simulate environmental stressors to better understand how long modules will perform in the field. While this process might reveal overall power loss, it doesn’t immediately reveal the specific causes behind the power loss. This makes it challenging to estimate the system lifetime and makes it difficult to compare various module types. Just like photography was once used to settle the infamous debate about whether a horse is airborne when it gallops, periodic photos taken over the course of hundreds of hours can be used to illuminate the precise areas of a solar cell that have degraded and are contributing to power loss.

solar case western reserve

With support from the Solar Energy Technologies Office, Case Western Reserve University is illuminating the degradation process. Through electroluminescence (EL) imaging, the research team is taking pictures of the cells throughout the module degradation testing process. The exposed images are then strung together over time and each image can be classified by degradation type, including cracked cells and busbar and gridline corroded cells. Busbars and gridlines or “fingers” span the length and width of the top contact of a solar cell and work together to transport electric current and minimize power losses. The fingers collect current and feed into the busbars, which help to deliver current that’s then used to generate electricity. Mitigating cracks and damages to these areas of a solar cell can help to build hardier solar cells, which translates to solar panels that last longer at lower costs.

Image data can reveal a lot about the health of the module and when images are captured through time, they can be used to quantitatively evaluate the progression of changes over the module’s lifetime. However, it can take researchers many hours to manually view cell images and classify them. To address this challenge, the research team pairs the evolution of EL images with machine learning to automatically find cracked cells, busbar corrosion, and gridline corrosion during module durability tests. The resulting algorithm is trained to recognize thousands of cell images in a myriad of degradation states and can classify 14,200 cell images with 98.7% accuracy in under two minutes.

This method can be used to analyze how cracks and corrosion change over time and will be able to pinpoint when, where, and how the cells got damaged. Once the images have been classified, the researchers use them to correlate patterns of power loss with specific degradation mechanisms.  Researchers have found a direct relationship between modules that are classified as having no corrosion, minor corrosion, or major corrosion near the busbars and the amount of power loss that’s observed. After automatically classifying module types, the algorithm was able to show that modules with major corrosion produce about two-thirds of the power of modules with little corrosion, primarily due to the electron carrier losses near the busbars.

case western reserve solar

A PV module exposed in damp heat for 500 hours (left) is compared to the same module (right) after 3,000 hours of exposure, showing substantial busbar corrosion. Image courtesy of Case Western Reserve University.

This automated classification system is brand agnostic and  can be used to improve labeling consistency and remove subjectivity in the “pass vs. fail” module tests used to rate solar modules on the market today. This durability research ultimately benefits solar customers who want their solar panels to last for two to three decades and able to survive a major hail storm.

Manufacturers have already expressed interest in the EL machine learning platform. Case Western Reserve University researchers will further investigate how EL images can be used to identify how degradation forms differ between various types of silicon solar cells. The research team hopes to eventually open-source the work, enabling module reliability researchers across the world to benefit from the added insight. This will empower solar researchers to work toward creating solar panels that can last for 50 years or more, helping to further reduce costs and meet the DOE solar office’s 2030 solar cost targets.

Learn more about the Solar Energy Technologies Office’s photovoltaics projects.