PSI - Issue 13

Vit Krivy et al. / Procedia Structural Integrity 13 (2018) 825–830 Vit Krivy, Monika Kubzova, Katerina Kreislova, Martin Krejsa /Structural Integrity Procedia 00 (2018) 000 – 000

829

5

4. Probabilistic prediction model for calculation of corrosion losses after one year of exposure

Regression analysis and predictive intervals have been calculated for all the necessary environmental parameters, the course of which has changed significantly over the last decades. Each random variable was assigned to the corresponding probability density distribution . Probability analysis assumes a normal distribution of the observed climatic variables. In order to eliminate negative values, some environmental parameters assume a log-normal probability distribution. In addition, the created histograms are cut off in the value ±3 σ , see Marek et al. (1996). The characteristics of each distribution of randomly variable environmental parameters are given in Table 1. The source code for solving the above predictive models was programmed in the Delphi software. The code was then imported into the ProbCalc computation module as a DLL library , see Janas et al. (2017).

Table 1. Basic characteristics of input values – locality Kopisty.

Mean value Standard deviation S x

Variable

Distribution

Temperature T [°C]

parametric normal parametric normal

9.71

0.65 2.74 3.52 0.33 7.49

Relative humidity RH [%] Deposition rate of SO 2 [mg.m Deposition rate of Cl - [mg.m -2 .d -1 ]

77.23 12.64

-2 .d -1 ] parametric normal

parametric normal

1.08

Deposition of PM 10 [mg.m

-2 .d -1 ]

parametric log-normal

21.95

Rainfall RAIN [mm]

parametric normal parametric normal

447.64

116.63

Value of pH of rainfall [-]

5.81

0.57

The environmental characteristics, defined as random variables with corresponding probabilistic distribution, were used as the input variables for the calculation of histograms of corrosion loss r corr after one year of exposure. These histograms can then be applied as an input parameter to equation (1) for determining the long-term corrosion losses. The ProbCalc software was used to determine the corrosion loss histograms for selected prediction models according to EN ISO 9223, UN/ECE ICP and Multi-Assess program. Selected histogram of corrosion loss r corr after one year of exposure for locality Kopisty is shown in Figure 3 The obtained mean corrosion loss values r corr after one year of exposure (including standard deviations S x and coefficients of variation V x ) are given in Table 2.

Table 2. Values of corrosion loss r corr [µm] after one year of exposure.

locality Kopisty

locality Ostrava - Poruba

Analytical function and real test result

Mean v. m x

Std. dev. S x

Coeff. of var. V x

Mean v. m x

Std. dev. S x

Coeff. of var. V x

ISO 9223

31.68 45.52 58.33 21.40

4.87 4.93

0.15 0.11 0.18

18.21 32.21

3.66 6.92

0.20 0.21

UN/ECE ICP Multi - Assess

10.65

- -

- -

- -

Real test

-

-

The lowest values of corrosion losses r corr were determined by equation according EN ISO 9223; the highest values were predicted according formula from the Multi-Asses Program. The prediction models according to UN/ECE ICP and Multi-Assess seem to be too conservative when compared with the measured value from the atmospheric station Kopisty. To determine the most suitable model for prediction of weathering steel corrosion loss, more detailed analysis will need to be made based on comparing more atmospheric test results with corresponding prediction models. The result from the atmospheric station Kopisty indicates that the best-fitting prediction model for weathering steels is the relation given in EN ISO 9223, see also the comparison of real and calculated annual corrosion loss given in Kreislova et al. (2017). The values of the variation coefficient V x of the predicted corrosion loss histograms are within the range 0.11 till 0.21. The variability of predicted corrosion losses r corr is therefore relatively high, and it is inappropriate to use only one value ( e.g. the mean value m x ) for an accurate description of the expected corrosion loss. Probabilistic expression of the predicted corrosion loss is much more accurate.