PSI - Issue 57

Amaury CHABOD et al. / Procedia Structural Integrity 57 (2024) 701–710 Author name / Structural Integrity Procedia 00 (2019) 000 – 000

705

5

4. Probabilistic Fatigue 4.1. Understand uncertainties on loads

The general fatigue route involves a fine-tuned description of the fatigue properties of materials, geometry and loads applied to the component in service. There are two types of uncertainty, according Halfpenny, Bonato et al. (2019): - Reducible uncertainties (or epistemic uncertainties) - Irreducible uncertainties (or aleatoric uncertainties) Material uncertainties are inherent to the material’s microstructure and therefore irreducible, even with a large number of samples used to characterize fatigue properties. Load uncertainties can be reduced by improving knowledge of customer usage and environmental conditions. This article focuses on load uncertainties derived from customer usage. The massive amount of loading data in the data lake will be used to understand this uncertainty. One way of characterizing any random variable is the probability density function or PDF, using a distribution such as normal, log-normal, discrete, Weibull, or Rayleigh. The PDF can be applied to characterize the time series loading signal, with the mean value and standard deviation in the case of the normal distribution. Distribution fitting methods may include rank regression or maximum likelihood estimation (MLE), particularly suited to large samples and uncensored data. The PDF can also be applied to loading stored in a histogram format, instead of considering all the data points in the time signals. Indeed, a clever way of extracting relevant information for fatigue analysis is to construct histograms to build the mission profile, such as rainflow cycle count, FDS, RDS, or time at level, used by Chojnacki et al. (2019). The usage of a data lake provides a convenient way to search the database, retrieve events and such calculated histograms associated with parent events, and at the end to get mission profile or histogram PDF. As described in NF X 50-144-3 (2021), for each value of the histogram, the damage at each frequency for example, the PDF parameters can be adjusted, and from a percentile value p=1-  , we can derive a design-focused value of FDS  . The way to calculate these FDS  is described in Table 1.