Cost-effective, fast, and nondestructive on-site characterization of photovoltaic plants is required to determine countermeasures against power loss, defects, or safety problems. Methods with small impact on the operation and a high throughput, such as infrared thermography (IR), or methods with high resolution for detailed defect information, such as electroluminescence (EL) imaging, are expedient. To combine high resolution and high throughput, we propose to use photoluminescence (PL) as an outdoor characterization method for full-sized module imaging. With PL imaging as with IR imaging, no electrical contact is necessary, yet image resolution is on par with EL images. Our outdoor PL setup features an excitation source with 18 broadband, white, high power, chip on board LEDs coupled with low-cost short pass filters. This setup is suitable for indium gallium arsenide and silicon detectors. Here, we compare the visibility of common defects, including short-circuited bypass diodes, cracks, potential induced degradation, snail trails, ethylene-vinyl acetate degradation, and interconnection failures, in PL images to that in state-of-the-art imaging techniques IR, EL, and ultraviolet fluorescence. We find that out of these seven defects, five can be detected well, cell cracks under certain conditions, and interconnection failures not at all. We discuss how different techniques are complementary to enable better defect detection.

Photoluminescence for Defect Detection on Full-Sized Photovoltaic Modules

Failures of photo

Energy losses of photovoltaic (PV) plants because of soiling are a problem in all regions, including Germany. Soft soiling, caused by a uniform dust film, is shading a PV module and is reducing yield depending on the thickness of the debris layer. Hard soiling, caused by agglomerations of dust or dirt, only covers parts of a PV module, and it causes local shading. A spot check of modules from different PV power plants in Germany revealed a power reduction because of soft- and hard soiling of up to 6%. Determination of soiling type and amount is a prerequisite for decisions about cleaning and for optimizing yield. In this article, we show that luminescence imaging can be used to detect and characterize soiling and quantify losses. For this purpose, we compare luminescence images before and after cleaning and we show that soiling becomes detectable and power losses quantifiable from difference images. We also show that the impact of hard soiling can be quantified from a single photoluminescence (PL) image. This technique may enable a fully automated quantification of power losses because of hard soiling in PV- modules and strings in the future. 

Luminescence Analysis of PV-Module Soiling in Germany

Classification machine learning models require high-quality labeled datasets for training. Among the most useful datasets for photovoltaic array fault detection and diagnosis are module or string current-voltage (IV) curves. Unfortunately, such datasets are rarely collected due to the cost of high fidelity monitoring, and the data that is available is generally not ideal, often consisting of unbalanced classes, noisy data due to environmental conditions, and few samples. In this paper, we propose an alternate approach that utilizes physics-based simulations of string-level IV curves as a fully synthetic training corpus that is independent of the test dataset. In our example, the training corpus consists of baseline (no fault), partial soiling, and cell crack system modes. The training corpus is used to train a 1D convolutional neural network (CNN) for failure classification. The approach is validated by comparing the model’s ability to classify failures detected on a real, measured IV curve testing corpus obtained from laboratory and field experiments. Results obtained using a fully synthetic training dataset achieve identical accuracy to those obtained with use of a measured training dataset. When evaluating the measured data’s test split, a 100% accuracy was found both when using simulations or measured data as the training corpus. When evaluating all of the measured data, a 96% accuracy was found when using a fully synthetic training dataset. The use of physics-based modeling results as a training corpus for failure detection and classification has many advantages for implementation as each PV system is configured differently, and it would be nearly impossible to train using labeled measured data.

https://www.mdpi.com/1996-1073/15/14/5085/pdf?version=1657631330

Physics-Based Method for Generating Fully Synthetic IV Curve Training Datasets for Machine Learning Classification of PV Failures

This article presents image processing methods for reconstructing and calibrating contactless electroluminescence (EL) images acquired with laser line scanning for diagnostic of photovoltaic modules. The method specifically focuses on two outputs: a contactless conventional EL that is calibrated to correspond to electrically biased EL and a contactless highlighted EL that emphasizes visualization of defects. Results are evaluated first by visual inspection, showing similar resolution to electrically biased EL at 2% or 5% Isc. A quantitative analysis by peak signal to noise ratio (PSNR) measures also assesses the resemblance between the reconstructed contactless conventional EL and the reference electrically biased EL.

/pdf/reconstruction-and-calibration-of-contactless-1q4n0yw0.pdf

Reconstruction and Calibration of Contactless Electroluminescence Images From Laser Line Scanning of Photovoltaic Modules

Deep-learning-based pipeline for module power prediction from electroluminescense measurements

Automated inspection plays an important role in monitoring large-scale photovoltaic power plants. Commonly, electroluminescense measurements are used to identify various types of defects on solar modules but have not been used to determine the power of a module. However, knowledge of the power at maximum power point is important as well, since drops in the power of a single module can affect the performance of an entire string. By now, this is commonly determined by measurements that require to discontact or even dismount the module, rendering a regular inspection of individual modules infeasible. In this work, we bridge the gap between electroluminescense measurements and the power determination of a module. We compile a large dataset of 719 electroluminescense measurementsof modules at various stages of degradation, especially cell cracks and fractures, and the corresponding power at maximum power point. Here,we focus on inactive regions and cracks as the predominant type of defect. We set up a baseline regression model to predict the power from electroluminescense measurements with a mean absolute error of 9.0+/-3.7$W_P$ (4.0+/-8.4%). Then, we show that deep-learning can be used to train a model that performs significantly better (7.3+/-2.7$W_P$ or 3.2+/-6.5%) and propose a variant of class activation maps to obtain the per cell power loss, as predicted by the model. With this work, we aim to open a new research topic. Therefore, we publicly release the dataset, the code and trained models to empower other researchers to compare against our results. Finally, we present a thorough evaluation of certain boundary conditions like the dataset size and an automated preprocessing pipeline for on-site measurements showing multiple modules at once.

Luca Reeb

Papers

Deep-learning-based pipeline for module power prediction from electroluminescense measurements

Deep Learning-based Pipeline for Module Power Prediction from EL Measurements.

Deep Siamese Metric Learning: A Highly Scalable Approach to Searching Unordered Sets of Trajectories