A Procedure for Verification of Models used to predict Chloride Ingress into Concrete
Paper i proceeding, 2006
There are several mathematical models available which predict chloride ingress into concrete. A requirement for all these models is that the predictions both have good precision, i.e. have small scatters, and accuracy, i.e. correspond well with measured data. In this paper a procedure to investigate the precision and accuracy of prediction models, used to predict chloride ingress into concrete, is described. The procedure involves determination of the areas, which the measured and predicted profiles, including their scatters, cover in chloride ingress diagrams. The uncertainties in the measured profiles depend on variations in concrete properties, exposure conditions and uncertainties in the measuring techniques, while the uncertainties in the predicted profiles depend on the uncertainties of the input data in the models. The precision of the models is evaluated by studying the areas of the predicted profiles. If the areas are small the precision is good and vice versa. The accuracy of the models can be evaluated by comparing the areas where the measured and predicted profiles are overlapping each other with the total area of the measured and predicted profiles. The quotient between the overlapping and total areas gives a picture of the accuracy where large and small quotients mean good and bad accuracy respectively.
Three predictions models are evaluated with the proposed procedure, namely the ClinConc model (physical model), DuraCrete model (empirical model) and the Error-function model (empirical model). The evaluation has been made in four different exposure conditions; marine submerged, splash and atmospheric conditions and along thaw-salted roads. The verification has been made both for short and long term exposure. The results from the verification are combined into success ratios by multiplying them with weight factors. The weight factors are used to show the relevance of each verification, where verifications made in marine submerged conditions and with long term data are considered more relevant than the other verifications. The final results show that the ClinConc model gives both better precision and accuracy compared with the other models. There were large variations in the accuracy between the different exposure conditions, where the best accuracy for all models was achieved in marine submerged conditions. An explanation to this observation is the difficulties to define appropriate boundary conditions in exposure conditions where the concrete is not constantly exposed to seawater.
Finally examples are given service life designs made with the evaluated models for concrete exposed in marine submerged and atmospheric conditions. The service life is defined to be ended when reinforcement corrosion is initiated. The predictions have been made with both mean value and pessimistic approaches, where in the latter all influencing parameters are put in such way that a lower limit of the service life is achieved, to illustrate how uncertainties influence service life predictions. The results show that there are significant differences between predictions made with mean value and pessimistic approaches. This is especially the case if the predicted profiles are flat. The predictions made with the empirical models can, however, be questioned since these are made far beyond the data the models are verified against.
reinforcement corrosion marine environment