PSI - Issue 84

Alessandro Nettis et al. / Procedia Structural Integrity 84 (2026) 653–660

659

for the obtained dataset, i.e., the right-skewed Gumbel distribution and the Johnson’s S U distribution. They both have satisfied the Kolmogorov–Smirnov (KS) and Cramér–von Mises (CVM) goodness-of-fit criteria (p-values greater than 0.05). It is worth mentioning that the absolute best fit is given by the Johnson’s S U distribution (KS p-value equal to 0.93 and CVM p-value equal to 0.93, compared to the Gumbel distribution which showed KS p-value equal to 0.75 and CVM p-value equal to 0.67). The Johnson’s S U distribution is a very flexible four-parameter distribution and can easily fit asymmetric data with heavy tails. However, other considerations must be taken into account to choose the best distribution. Indeed, the original dataset includes data regarding the maximum pit detected on each wire. This can be interpreted as an extreme event among the entire set of pit detected on a single wire. As a consequence, the distribution of -parameters derived from such a built dataset represents a distribution of extreme events. In such conditions, the extreme value theory can be applied and the distribution converges to a Generalised Extreme Value (GEV) distribution, and the Gumbel distribution belongs to this family, as it is a Type I GEV distribution. For this reason, it is assumed as the most plausible. The parameters of the Gumbel distribution describing the parameter are reported in Table 2, and a graphical presentation of the distribution and the histogram of the dataset is provided in Figure 4.

Figure 4. Gumbel distributions, best fitting for the dataset. Table 2. Parameters of the Gumbel distribution describing the parameter . Loc Scale Mean Variance

Standard Dev.

Skewness 1.139547

Kurtosis

Coeff. of Variation

0.056105

0.425311

0.301602

0.297551

0.545483

2.4

1.80862

5. Conclusions This paper presents a novel geometric modelling strategy suitable for prestressing strands affected by pitting corrosion. Chloride-induced corrosion can affect prestressing strand surface in highly localised areas, referred to as corrosion pits. The proposed modelling strategy aims to model the steel area loss in a steel strand on a wire-by-wire basis through a parabolic shape intersecting the steel wire cross-section. A parabolic model is defined by two parameters, i.e., the maximum pit depth and the parabola opening. The two parameters uniquely define a parabola, and the area resulting from the intersection of the parabola and the wire circumference defines the area loss. The model has been calibrated on a dataset available in the scientific literature of naturally corroded strands extracted from 10-years-old PC beams (Vecchi et al., 2021). The dataset describes a sample of 24 corroded 7-wires strands. From this dataset, the values of the area loss and the maximum pit depth have been extracted for the pit with the maximum area on each wire of the sample. The parabolic model has been applied to each wire, starting from the maximum pit depth and calibrating the value of the parabola opening such that the area resulting from the intersection between the parabola and the wire circle matched the one detected from the experiments. As a result, a sample of values of parabola openings has been obtained. A statistical distribution has been fitted to these data, and the Gumbel distribution (Type I GEV distribution) has been found to provide the best fit to the considered dataset. The proposed model matches well with other types of classification previously proposed in the literature for corrosion pit morphology (Jeon et al., 2019). The obtained Gumbel distribution is consistent with extreme value