Benchmarking Power Loss Simulation Models for Cracked Photovoltaic Cells Using Electroluminescence Images: The Effect of Daylight and Image Resolution

Algorithms and models for simulating power loss in photovoltaic (PV) cells using electroluminescence (EL) images are typically developed, trained, and validated on high-resolution images acquired under dark laboratory conditions. In this work we benchmark the performance of an analytical model (bELMO) and a data-driven power loss simulation model (DTU ML) under laboratory and field imaging conditions. Our goal is to evaluate the impact of solar illumination noise, camera type, and image resolution on the accuracy of power loss estimation. EL images were acquired from cells with varying numbers of busbars. The dataset includes both pristine (defect-free) cells and cells exhibiting cracks of varying severity. Laboratory imaging was conducted using both a CMOS camera and a lower-resolution InGaAs camera to assess the models’ robustness to changes in resolution. Additionally, EL images were acquired under daylight conditions to evaluate the models’ performance with field-acquired images. Results show that both models perform well for cells with low power loss, regardless of image resolution or lighting conditions. However, for more severely degraded cells, both models exhibit increased error under both dark and daylight conditions. High-irradiance daylight EL images led to reduced modeling performance for both approaches, though bELMO consistently achieved lower average error across all conditions.