PREDICTING LONG-TERM COATING SYSTEM PERFORMANCE USING NEURAL NETWORK MODELS AND SENSITIVE SHORT-TERM COATING MEASUREMENTS

The coatings industry is heavily dependant on exposure testing to develop, qualify and verify coating protectiveness. Protective properties of coatings degrade slowly in most exposure test protocols. Besides being too slow for modern patience levels, coating degradation is too slow for most materials and process development programs, and is usually too slow for diagnosing unexpected field failures. Any attempt to provoke an accelerated corrosion response by increasing the aggressiveness of the exposure comes with the uncertainty that the fundamental nature of the damage mechanism has been changed to one that is irrelevant to the application. One solution to this conundrum is to couple a highly sensitive diagnostic method for coating degradation with exposure testing and determine the relationship between the earliest stages of coating degradation and the ultimate occurrence of coating failure. We have correlated the relationship between the earliest stages and last stages of coating degradation by relating electrochemical spectroscopy (EIS) data collected after exposure times as short as 24 hours to results from visual inspection made after 720 hours. This has been done using neural network models. Generic model frameworks have been tailored for this purpose using supervised learning methods. Trained models recognize the subtle and complex relationships between EIS data sets captured after short exposures and the results of exposure tests captured after long exposure times. Once trained and validated, neural network models become predictive tools capable of significant foreshortening of long exposure test protocols. Sensitivity analyses applied to these neural network models allow mechanistic insights to be developed to help understand how and why various coating systems fail the way they do. These ideas will be presented and addressed through examples involving inorganic and organic coating systems.