Introduction

Structural Health Monitoring (SHM) is essential for maintaining the reliability and long-term performance of photovoltaic (PV) systems, particularly in regions exposed to harsh environmental conditions. Photovoltaic installations in Nigeria operate under high solar irradiance, elevated ambient temperatures, dust accumulation, and limited maintenance accessibility, which accelerate structural degradation such as cell cracking, delamination, thermal hotspots, and shading losses. Existing SHM research is largely dominated by laboratory-based electroluminescence (EL) inspection and single-metric performance evaluation, limiting scalability and field applicability. To address these limitations, this study proposes a structured AI-driven multi-modal SHM evaluation framework that integrates unmanned aerial vehicle (UAV) imaging, thermal and RGB sensing, and deep learning diagnostics. Unlike experimental studies focused on single datasets, this work develops a systematic analytical framework that synthesizes and evaluates published photovoltaic diagnostic pipelines using objective multi-criteria decision analysis, enabling consistent cross-study comparison under field-relevant deployment conditions.

Methods

A literature-supported analytical framework was developed using Multi-Criteria Decision Theory (MCDT) to evaluate competing AI–hardware SHM pipelines. Eight representative AI–hardware alternatives were selected from peer-reviewed photovoltaic diagnostic studies using strict inclusion thresholds (≥90% Accuracy/F1/AUC or low regression error metrics) and verified field applicability. Performance indicators included Accuracy, F1-score, Area Under the Curve (AUC), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and coefficient of determination (R²). To ensure methodological consistency across heterogeneous studies, performance metrics were normalized and objectively weighted using the CRITIC (Criteria Importance Through Intercriteria Correlation) method. Robustness-sensitive criteria received the highest weights (AUC = 0.2782; RMSE = 0.1978; MAE = 0.1835; R² = 0.1844), while Accuracy received a minimal weight (0.0071) due to uniformly high reporting across studies. The weighted decision matrix was subsequently ranked using the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) to identify field-optimal SHM pipelines.

Results

TOPSIS ranking identified CNN combined with thermal imaging as the most effective SHM solution (Closeness Coefficient Ci = 0.567), followed by CNN + UAV (RGB/Thermal) (Ci = 0.505) and Hybrid CNN + Sensors/IoT + UAV systems (Ci = 0.497). CNN-thermal pipelines consistently demonstrated high diagnostic performance (≈95–96% accuracy) and strong AUC values (~0.96), while maintaining operational scalability for large photovoltaic farms. UAV-enabled CNN frameworks further improved inspection coverage by enabling non-contact, large-scale monitoring of distributed PV arrays. In contrast, EL-based CNN methods ranked lower (Ci = 0.370) due to laboratory constraints, while classical machine-learning approaches such as SVM and ensemble tree models exhibited weaker generalization and incomplete metric reporting (Ci = 0.032–0.258). Hierarchical clustering analysis further confirmed that UAV-thermal CNN pipelines form a high-performance family of SHM solutions optimized for real-world deployment environments.

Conclusion

This study provides a methodologically consistent, literature-derived multi-criteria evaluation framework for AI-based photovoltaic structural health monitoring pipelines. The results indicate that UAV-mounted thermal CNN workflows offer the most robust and scalable SHM solution for photovoltaic installations operating under Nigerian environmental conditions. By integrating multi-modal sensing with objective decision analysis, the proposed framework enables systematic selection of field-deployable PV diagnostic technologies and bridges the gap between laboratory-based research and practical solar farm monitoring in sub-Saharan Africa. The framework also establishes a foundation for future experimental validation using field-acquired UAV and thermal datasets.