Akademik Veri Yönetim Sistemi
5TH INTERNATIONAL CONFERENCE ON INNOVATIVE ACADEMIC STUDIES ICIAS 2024, Konya, Türkiye, 10 - 11 Ekim 2024, ss.613-618
Fault diagnosis in photovoltaic (PV) panels is
crucial to ensure efficient energy generation, reduce maintenance costs, and
enhance system reliability. Convolutional Neural Networks (CNNs) are widely
employed in classification tasks due to their ability to automatically extract
and learn meaningful features from image data, making them an ideal choice for
identifying faults in PV panels. In this study, we utilize the publicly
available Electroluminescence Photovoltaic (ELPV) dataset, which consists of
2624 grayscale images of solar cells, each with a resolution of 300x300 pixels.
The dataset includes functional and defective cells, and each image is
annotated with a defect probability value ranging from 0 to 1, along with
information on the crystalline type of the solar module (mono- or
polycrystalline). Based on these defect probabilities, we categorized the
images into four classes: defective_mono, not_defective_mono, defective_poly,
and not_defective_poly. To classify these classes, we designed a custom 2D CNN
architecture tailored to effectively handle the specific characteristics of the
ELPV dataset. Through this approach, our model achieved a classification
accuracy of 82.51%. The results demonstrate that our CNN-based model can
reliably distinguish between defective and non-defective solar cells across
different crystalline types, providing a valuable tool for automated PV panel
inspection and defect detection. This work highlights the potential of deep
learning models to enhance the quality and performance of PV systems by
enabling precise and efficient defect classification.