Reliability Analysis of Photovoltaic Modules Based on Joint Bayesian Model

We develop a unified framework for forecasting the reliability of photovoltaic modules by integrating accelerated laboratory degradation measurements with field failure data. First, power-vs-time measurements at multiple temperatures are processed to estimate degradation rates via linear regression, and outliers are removed based on ±1 standard deviation rule. Rates are projected to service conditions using the Arrhenius law. Next, we introduce a simple transfer-learning calibration that maps laboratory projections to field observations via a linear log-rate adjustment, accounting for domain mismatch. Finally, we formulate a joint Bayesian model that simultaneously fits the Arrhenius parameters and the field failure times, yielding coherent posterior distributions for activation energy, pre-exponential factor, and failure-time variability. Posterior predictive inference provides credible intervals for key lifetime metrics. A numerical example demonstrates the methodology on real lab and historical datasets, showing that the Bayesian joint model closely matches empirical lifetimes while properly quantifying uncertainty. This approach offers a principled pathway for combining accelerated testing and operational data in photovoltaic reliability assessment.