An artificial neural network approach to stimuli-response mechanisms in complex systems

Artificial intelligent methods in the form of neural networks are applied to a set of different families of architectural alkyd enamels as a rational approach to the study of complex stimuli-response mechanisms: 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. It is shown that good predictive power (of over 90%) is obtained within given appropriate uncertainty tolerances, and that the set of variables which most significantly affect a given property can also be determined through this methodology via a sensitivity analysis.