Researchers Estimate PV Power Loss from A Single Luminescence Image

Machine learning identifies the dominant degradation mechanism, while a physics-based model converts one luminescence image into power-loss maps from cell to string level

A drone records electroluminescence images at night, allowing the method to map estimated power loss at cell, module and string level across a PV plant. (Image Credit: Peng et al., Matter & Light)

Published on:

Sep 29, 2026, 7:43 am

Key Takeaways

The method needs one electroluminescence image per module, rather than images taken under several electrical conditions
For more than 300 field-retrieved modules, reconstructed power showed a root-mean-square error of 0.5% against direct current-voltage measurements
Nighttime drone imaging at a separate 186 kW PV plant produced estimated loss maps at cell, module and string level

Electroluminescence images reveal dark areas associated with damage or degradation in a PV module, but a dark area alone does not quantify the resulting power loss. Conventional quantitative analysis usually requires several images taken under different electrical conditions, slowing inspections across large PV plants.

To address this gap, researchers from Zhejiang University, the University of New South Wales (UNSW), and Quantified Energy have developed a method to estimate power loss from one luminescence image per module. A lightweight machine-learning classifier identifies whether the dominant loss comes from cell recombination or from increased electrical resistance caused by defects such as cracks and broken grid lines. A physics-based solar-cell model then reconstructs the module's power output and spatial loss distribution. The method uses an electrical measurement from one reference module to calibrate image brightness for other modules in the same batch.

The researchers first validated the underlying diagnostic model . They tested more than 300 crystalline-silicon modules retrieved from PV plants. They reconstructed each module's power from a single image and compared the result with a current-voltage measurement. The reported root-mean-square error was 0.5%, with most module-level errors within ±1.5%.

The team also subsequently demonstrated the method using nighttime drone-based electroluminescence imaging at a separate 186 kW PV plant with 342 modules. For a representative group of 24 modules, the method estimated an average power loss of 7.53%, while localized degraded regions showed losses approaching 25%. The researchers also compared 3 strings of 28 modules affected by different degradation mechanisms. However, the study does not report an independent comparison between these drone-derived estimates and measured power values from the plant.

The researchers say the loss maps could help plant operators prioritize closer inspection, repair or replacement.

The study, One-Shot Luminescence Diagnostics For Field-Scale Photovoltaics, was published online in Matter & Light.

University of New South Wales (UNSW)