Data-analysis method for material optimization by forecasting long-term chemical stability

The development of material including optimization of long-term effects often requires accelerated testing methods or calculation of future material behavior to avoid excessive project time. As controlled conditions with few parameters allow the application of laws, more complex modeling fails if the law does not cover all parameters or when several simultaneously occurring events cannot be combined to one comprehensive model. This approach describes a chemometric method accelerating the material development by forecasting the material behavior based on similar realizations under same test conditions. The capability of this method was analyzed with a synthetic data set simulating a typical application case including noise and the study for reducing the thermal yellowing of a coating. The thermal yellowing proceeds slowly over several weeks due to chemical reactions of the polymer influenced by synergistic effects of the coating ingredients. Moreover, it was shown how this forecasting method can be combined with experimental design via Gaussian process regression and Bayesian optimization. The comparison of the model based on the forecasts versus the observed values was shown drawn from the results of the start experiments. Furthermore, the suggestions for the next adaption based on forecasted values were evaluated.