Atmospheric corrosion: statistical validation of models

Abstract

In this paper we discuss two different methods for validation of regression models, applied to corrosion data. One of them is based on the correlation coefficient and the other one is the statistical test of lack of fit. Both methods are used here to analyse fitting of bilogarithmic model in order to predict corrosion for very low carbon steel substrates in rural and urban-industrial atmospheres in Uruguay. Results for parameters A and n of the bilogarithmic model are reported here. For this porpose, all repeated values were used instead of using average values as usual. Modelling is carried out using experimental data corresponding to steel substrates under the same initial meteorological conditions (in fact, they are put in the rack at the same time). Results of correlation coefficient are compared with the lack of it tested at two different signification levels (a= 0.01 and a= 0.05). Unexpected differences between them are explained and finally, it is possible to conclude, at least in the studied atmospheres, that the bilogarithmic model does not fit properly the experimental data.