A reliability-based methodology has been successfully employed in mathematically linking field and laboratory exposure results and in predicting the outdoor performance of a neat epoxy coating.

Laboratory aging tests were experimentally designed and exposed to wide range of extremely well-controlled temperature, humidity, spectral waveband, and spectral intensity conditions. Designed laboratory experiments were conducted in the NIST SPHERE, a 2-meter integrating sphere in which

1) each of above stated exposure variables were precisely, accurately, and independently controlled over long exposure periods and

2) all known sources of experimental error common to commercial exposure equipment have been eliminated and no new error sources have been introduced.

 The laboratory experiments were specifically designed to validate the total dosage model, reciprocity law, and the additivity law, over wide temperature and relative humidity ranges. For the studied material, all three models appear to be valid. Epoxy coated specimens were exposed to 4 relative humidities, 4 temperatures, 4 ultraviolet wavebands, and 4 irradiance levels for each waveband. Numerous analytical chemical and physical measurements of degradation were taken on each specimen at numerous times throughout their exposures. Extensive efforts were made to ascertain the underlying failure mechanisms.

 Outdoor exposure experiments were conducted on the NIST roof located in Gaithersburg, MD. This exposure site was instrumented to characterize the field exposure environment in exactly the same manner as the laboratory exposure environments. Specifically, the outdoor site monitored or estimated solar spectral irradiance, panel temperature, and panel moisture content every 12 minutes of every day. Field experiments were started at the beginning of each of 18 months over a three year period. Degradation of the exposed specimens was characterized in the same manner as the specimens exposed in the laboratory. Degradation data for every specimen was collected numerous times for each experiment.

 Field and laboratory exposure results were linked by their failure mechanisms and via the total dosage model. Two cumulative damage models were developed to predict field performance—an analytical model and a neural network model. Both models assumed only that the total dosage and additivity law were obeyed and that the reciprocity law or a variant of the reciprocity law was valid.

 Model parameters were estimated from laboratory exposure data. Estimates of field response for all 18 experiments were made from these parameterized models and from field exposure data for each experiment. Predictions from both models were excellent. The neural network model, being a data based model, predicted field degradation response slightly better than did the analytical model, but both estimates were well-within 10% of the actual field response data for all 18 field experiments.

 It is concluded that the reliability-based methodology is capable of linking laboratory and field exposure data and of predicting the performance of coated specimens exposed in the field.