Design of Experiment for Degradation Modelling of PV Modules

Due to the large effort in terms of time and equipment necessary for reliability testing of photovoltaic (PV) materials, components and modules, the PV community has always endeavored to obtain service life estimates, based on an extrapolation of measurement and characterization data from accelerated aging tests or modelling. One major focus in recent years has been to link degradation of PV module materials with the observed power loss, based on a better understanding of the complex underlying degradation mechanisms and material interactions. In general, the main challenge for modelling of degradation behavior is to provide enough samples and perform appropriate aging tests to enable meaningful modeling while still maintaining a manageable test schedule. The main aim of the paper is to derive an optimum design of experiment based on the experiences made in the Austrian flagship projects Infinity and ADVANCE!. Re-evaluation of the existing data revealed which experimental data were most useful for modeling. Several critical issues were identified that significantly complicated modelling and reduced the validity of the results. The observed issues range from data availability, suitability of aging tests, comparability of characterization methods to relics from specimen preparation causing premature failure. For example, inconsistent electrical behavior has been noted for modules of same design but with manufacturing dates several months apart. These inconsistencies could be attributed to handling or device inaccuracies, different material batches or shelf life of certain components. Missing data points often were a result of logistical problems, where samples got lost or broken during transport between partners. An optimum design of experiment which generates the necessary input data for meaningful degradation modelling of PV modules needs to focus on (1) Reproducible sample preparation (2) Suitability of accelerated aging tests with proper test intervals (3) Suitability of characterization methods, which in the best case allow for predictive maintenance measures (4) Appropriate data preparation allowing for automated feature extraction.