Optimizing Service Life Prediction Models of External Paint Finishes

Abstract

In the present research effort, the topic of service life prediction is revisited, using external paint finishes on rendered facades as a case study. A statistically robust approach to develop ordinary linear regression models is proposed, covering the identification of the explanatory variables to include in the model and the identification of outliers. The application of this approach to the analyzed case study improved the determination coefficient (R2) up to 0.84, from the 0.74 obtained in the original research in which the data set was presented. In addition to the increase in accuracy, the novel approach proposed enabled the identification of four factors having a statistically significant influence on the degradation rate of the paint finishes: (1) the urbanization density; (2) the humidity level; (3) the type of surface finishing; and (4) the facade orientation. Finally, to enable the development of a hybrid model, a novel approach is suggested whereby the model uncertainty is used to combine statistical and expert inputs. The hybrid modeling approach was proposed to allow tuning the model performance in situations in which experts can identify specific characteristics or conditions affecting the service life not taken into account by the statistical model.