Article Per Year (5 Year)
p-Index From 2021 - 2026
0.23
P-Index
This Author published in this journals
All Journal International Journal of Renewable Energy Development
Dang, Thanh Nam
Unknown Affiliation
Control & Systems Engineering Earth & Planetary Sciences Electrical & Electronics Engineering Energy Engineering

Published : 1 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 1 Documents
Search

 1 
Towards self-diagnostic solar farms: Leveraging EfficientNet and class activation mapping for predictive maintenance Nguyen, Du; Nguyen, Thi Bich Ngoc; Nguyen, Duc Chuan; Chau, Thanh Hieu; Duong, Minh Thai; Dang, Thanh Nam
International Journal of Renewable Energy Development Vol 15, No 2 (2026): March 2026
Publisher : Center of Biomass & Renewable Energy (CBIORE)

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.61435/ijred.2026.62298

Abstract

The high rate of utility photovoltaic (PV) system development has increased the demand for stable, automated, and interpretable fault diagnostic systems that can be utilised in real-world environments. Solar farms with a large size are increasingly making conventional manual inspection methods impractical, and triggering the use of intelligent data-driven solutions. This paper presents a justifiable deep learning model for automated fault classification of solar panels based on the EfficientNet-B2 architecture combined with Gradient-weighted Class Activation Mapping (Grad-CAM). A six-class image dataset made of clean panels and five prevalent fault types is used. The two stages of transfer learning used to train the model include a warm-up phase and selective fine-tuning of upper network layers. Data augmentation is also performed extensively to make it more robust to changing illumination, viewing angles, and environmental noise. The experimental findings reveal consistent convergence and excellent generalization ability, and a high level of classification accuracy of all types of faults, as it achieved high classification accuracy, macro-averaged F1-scores exceeding 0.90 for most fault classes, and a macro-averaged ROC–AUC of approximately 0.981, highlighting the robustness and reliability of the proposed diagnostic model. The suggested structure will provide a scalable, interpretable, and realistic predictive maintenance of solar farms of the next generation with self-diagnostic capabilities.
Co-Authors Chau, Thanh Hieu Duong, Minh Thai Nguyen, Du Nguyen, Duc Chuan Nguyen, Thi Bich Ngoc