A Multimodal Data-Driven Framework for Failure Analysis and Performance Degradation of Photovoltaic Panels in Smart City Applications

Parladi E., Adıgüzel E.

IEEE ACCESS, cilt.14, ss.64666-64680, 2026 (SCI-Expanded, Scopus)

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 14
  • Basım Tarihi: 2026
  • Doi Numarası: 10.1109/access.2026.3683953
  • Dergi Adı: IEEE ACCESS
  • Derginin Tarandığı İndeksler: Scopus, Science Citation Index Expanded (SCI-EXPANDED), Compendex, INSPEC, Directory of Open Access Journals
  • Sayfa Sayıları: ss.64666-64680
  • İstanbul Üniversitesi-Cerrahpaşa Adresli: Evet

Özet

This study investigates the long-term aging effects and structural degradation mechanisms of photovoltaic (PV) modules integrated into smart city furniture, with a particular focus on panels of 10 W, 25 W, and 40 W nominal ratings. A multimodal, data-driven diagnostic framework is adopted by combining electroluminescence (EL) imaging, infrared (IR) thermography, and electrical parameter characterization based on current–voltage (I–V) and power–voltage (P–V) analysis. Machine learning–based data-driven models, including Random Forest, Extreme Gradient Boosting (XGBoost), and Multilayer Perceptron (MLP), are employed to determine the relative importance of diagnostic parameters and to evaluate sustainable computing implications. Accordingly, the study should be interpreted as a module-level diagnostic case study demonstrating how multimodal measurements and data-driven analysis can be integrated to investigate degradation patterns in field-aged photovoltaic modules. The experimental results reveal significant performance deterioration, with power losses of 37.5 % (40 W), 44 % (25 W), and 80 % (10 W) under operational conditions, further increasing under standard test conditions (STC). EL imaging identifies potential-induced degradation (PID), microcracks, soldering failures, and grid finger defects, while IR thermography confirms the presence of hotspot clustering and thermal anomalies strongly correlated with electrical losses. Machine learning analysis demonstrates that maximum power point voltage ( Vmpp ) and short-circuit current ( Isc ) consistently emerge as dominant predictors, while hotspot temperature differentials ( ΔT ) gain higher relevance in the MLP model, reflecting nonlinear thermal–electrical interactions. The findings underscore the potential of integrating multimodal diagnostics with energy-efficient, data-driven machine learning for real-time PV health monitoring in smart city applications. By reducing reliance on repeated physical inspections, the proposed framework supports sustainable computing, energy-aware diagnostics, and predictive maintenance strategies, thereby enhancing the resilience and operational efficiency of renewable energy infrastructures.