PSI - Issue 84

Silvia Manarin et al. / Procedia Structural Integrity 84 (2026) 231–239

237

Figure 2 (a) Boxplot of ϑ values for 7-wire strand configuration’s measurements (b) Regression plots for 7-wire strand configuration's measurements.

5. Outliers-Filtered Experimental Data: improving accuracy and reliability The removal of outliers is a standard practice and is typically performed during the saving phase of each measurement. As illustrated in Figure 3, the way the data points fit the proposed curve allows for an intuitive identification of anomalous values that negatively affect the accuracy of the stress calculation.

Figure 3 Identification of the outlier points to be removed

These points can then be excluded from the final analysis. In the example shown in Figure 3 — based on the GNR PVII algorithm — the two outermost points at both positive and negative ψ angles were removed, as they clearly deviate from the expected parabolic trend. Compared to the initial analysis, the refined dataset shows a closer alignment between the estimated and applied stress values. In particular, the box plots in Figure 4 exhibit a noticeable reduction in variability, with tighter interquartile ranges and fewer extreme values. The updated regression and box plot diagrams, obtained after the exclusion of outliers, confirm a significant improvement in the reliability of the XRD measurements. The regression diagrams also display a clearer linear correlation between the actual and estimated stress values. The measured values seem to overestimate the tensions at lower levels. Data points are now more tightly clustered around the ideal 1:1 line, especially for the 7-wire strand testing campaign, suggesting that the XRD