Answers are grounded in this paper and cite their sources.
Photovoltaic (PV) fault severity quantification from UAV-based infrared thermography increasingly relies on composite indices. This study evaluates the PV Fault Severity Index (PV-FSI) on the DTU Drone Infrared Thermography Dataset (640 acquisitions of 43 deliberately degraded modules, seven defect types), with module-level statistics as the primary evidence. First, the Thermal Entropy Index (TEI) explains little of the PV-FSI variance ( R² = 0.032 ); its module-level association with PV-FSI is weak ( ρ = -0.112 , $$p = 0.476$$ ), and a stratifier that excludes TEI does not support a severity-dependent sign reversal. Second, per-defect principal component analysis separates a TEI-dominated crack family from an HSI/TNI-dominated high-intensity family; family-adaptive weighting increases between-defect-type separability (module-level Kruskal-Wallis H from 20.59 to 33.65), the gain persists in most held-out module splits and a ratio proxy assigns the family with leave-one-out accuracy 0.953. Third, a defect-specific residual irradiance correction (PV-FSI IR ), fitted on the 468 acquisitions meeting IEC TS 62446-3, removes the pooled irradiance drift ( H from 31.72 to 3.76). Referenced to nameplate power, PV-FSI shows no significant within-type association with Pₘₚₚ ; it is therefore positioned as a thermal severity descriptor for inspection triage, complementing electrical characterization. The results come from a single site without healthy reference modules and motivate multi-site validation.
No takes yet. Share an insight, caveat, or question.
Kabir et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: