Modeling, prediction, and analysis of alkyd enamel coating properties via neural computing

The use of artificial neural networks (ANNs) in the modeling and prediction of alkyd enamel coating properties, as well as in the sensitivity analysis that can be performed between the properties and the different paint components, are described. A feedforward neural network with sigmoidal activation functions was used with a conjugate gradient algorithm to recognize the complex input-output relation between the paint properties and the formula components. We restricted the study to only two properties of alkyd enamel paints: gloss and drying time. A database of five different families of alkyd enamel paints, containing the different components of the formulations as well as process information, was used in this study. The results obtained show, within the expected uncertainty tolerance, that predictive power of more than 90% for these two properties can be achieved. A sensitivity analysis was also performed using ANNs, yielding the relative importance of the different components of the formulation on the properties of the enamel coatings, which agrees with experience for gloss but gives mixed results for the drying time (apparently due to the high uncertainties in the measurement of this property).