PSI - Issue 76

Luca Esposito et al. / Procedia Structural Integrity 76 (2026) 50–58

51

As emphasized in the literature by Le et al. (2020) and Sausto et al. (2022), a wide variety of fatigue crack initiation mechanisms in machined AM components contribute to increased lifetime dispersion, necessitating more advanced probabilistic models to accurately describe fatigue behavior. Recent studies have shown that machined specimens printed with the loading direction aligned with the building axis (Z orientation) show significantly higher scatter in fatigue life compared to other building directions Le et al. (2020); Esposito et al. (2025). This scatter has been correlated with the orientation and morphology of internal defects, particularly the distinction between porosity and lack of fusion, Wu et al. (2021). Fractographic analysis revealed that shorter lifetimes are typically associated with LoF defects, while longer lives are observed when fatigue initiates from porosity-type defects. This highlights the importance of considering not only the defect size, but also the shape and orientation of the defect relative to the applied loading. This work addresses the limitations of traditional fatigue analysis by introducing a statistically robust method capable of accounting for the di ff erent contributions of defect-related mechanisms to fatigue life scatter. The focus is on machined specimens, where surface roughness e ff ects are minimized, and the intrinsic influence of internal defects becomes more pronounced. An unusually high scatter in fatigue life was observed in machined samples built with the loading direction aligned to Z-orientation. This scatter, often dismissed as experimental variability, is here attributed to the co-existence of two distinct defect populations. Weibull-based statistical models have long been used in fatigue to describe lifetime distributions, capture the e ff ects of defect populations, and manage censored data such as runouts. In particular, Brandl et al. (2012b) applied Weibull statistics to assess the sensitivity of the fatigue resistance of AM components to process parameters. Furthermore, Weibull formulations have also been successfully employed in fracture mechanics to model the scatter in fracture toughness measurements, as demonstrated by Esposito et al. (2007), thereby providing a consistent probabilistic framework across di ff erent failure mechanisms and confirming its wide applicability Weibull (1951). We propose a new statistical fatigue analysis framework based on a bimodal log-Weibull distribution, calibrated via the Maximum Likelihood Estimation Method (MLEM), which di ff erentiates the fatigue influence of two primary defect types: porosity and lack of fusion or porosity coalescence. This model captures the distinct fatigue behavior associated with each defect class. In doing so, the approach enhances the reliability of fatigue life prediction and contributes to the development of defect-informed design criteria for AM components. The material investigated in this study is the AlSi10Mg alloy, processed by selective laser melting (SLM). All specimens were built with the loading axis (Z-axis) normal to the building plate and fabricated using identical process parameters and geometry to ensure comparability. Two post-processing conditions were examined. In the first condi tion, specimens were tested in the as-built state, without any surface machining or heat treatment (Z-as-built). In the second, specimens underwent surface machining to remove surface roughness and manufacturing-induced anomalies, but without subsequent heat treatment (Z-machined). This approach enables the specific assessment of the influence of surface condition and internal defects on fatigue performance. Additional details regarding sample geometry and printing process parameters are provided in Esposito et al. (2025). 2.2. Fatigue Testing Fatigue tests were conducted using an Amsler Zwick 250 vibrophore under sinusoidal loading at a constant stress ratioof R = 0 . 1 and a frequency of about 65 Hz. The number of cycles to failure was recorded for each specimen, with right-censored (runout) data also included in the analysis. A total of 10 fatigue tests were performed on Z-machined specimens and 9 tests on Z-as-built specimens, with two runouts recorded in each set. The goal was to characterize and model the variability in fatigue lifetime for Z-oriented samples in both surface conditions. 2.3. S-N median curve 2. Materials and Methods 2.1. Material and Sample Preparation

As an alternative to the logarithmic linear model of the standard ASTM-E739 (2015), the following non-linear asymptotic formulation for the median fatigue life curve is proposed: