Severity-Graded, Step-Feature-Aware I-V Fault Diagnosis for PV Arrays: A Unified Rules + ML Pipeline

Photovoltaic arrays experience circuit faults (open/short), power-degradation faults (aging and potential-induced degradation), and current-mismatch faults (partial shading, hot spots, and cracking), each leaving a characteristic signature on current-voltage curves. We present a hybrid, multi-stage workflow that combines physics-informed features and rule-based logic with a lightweight learning back-end to deliver accurate, interpretable, and severitygraded diagnosis suitable for operations and maintenance. Preprocessing includes gap filling, outlier suppression, denoising, normalization by open-circuit voltage and short-circuit current, and irradiance/temperature compensation for fair comparison across modules. A step-aware segmentation module detects concave regions and step-like depressions associated with bypass-diode activation. We extract electrical descriptors (open-circuit voltage, short-circuit current, maximum-power-point voltage/current, and fill factor) and shape cues (end-region slope, step presence and depth, local goodness-of-fit around the step, and plateau slope). Diagnosis proceeds hierarchically: circuit-fault screening with severity grading, degradation identification via fill-factor reduction and end-region slope changes, and mismatch-subtype triage using plateau behavior and step-related current loss; when residual ambiguity remains (notably shading versus hot spots), a compact neural classifier provides a tie-break while preserving interpretability. Using the NIST photovoltaic monitoring dataset (six arrays; 2015-2018; minutelevel sampling with current-voltage sweeps, irradiance, and temperature), we curate 167 illustrative traces for validation and demonstrate robust multi-fault coverage, fine-grained mismatch triage, and clear severity grading.

Keywords: Photovoltaics; I-V curve; step features; fault diagnosis; severity grading.