PSI - Issue 57

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

708

8

analysis is performed using nCode DesignLife 2022.0. The assumption of linear behavior and high-cycle fatigue leads to the use of the Basquin SN curve and the linear accumulation of damage, in accordance with Miner's rule. By repeating the simulation for any set of random scaleFx and scaleFy values, according to their PDF form with the corresponding mean and standard deviation previously calculated (see Table 3), uncertainty is propagated into the simulation result, in our case the predicted fatigue life. The number of simulations is 30 for the current application. The advantage of a data lake environment is that the PDF parameter results are adjusted according to the user's request (here: model=Berline and motor=Electric). The single deterministic simulation using the mean values of scaleFx and scaleFy from Table 3 gives a lifetime of B50=5.17e8 cycles.

Fig. 4. Overview of the Monte Carlo simulation process

The advantage of using a data lake is that random values can be easily generated and all batch simulations run efficiently on the server. Any parameter in the analysis stream can be considered a random variable. This doesn't avoid any scripting operations, but the Python routines are integrated into the application process and hidden from the user, making deployment easier compared to the complexity of a Monte Carlo process. Indeed, an app is easily automated using an external call with a web URL, in which any variable and corresponding value are included. Next, we obtain the values of the damage results for any pair of input values (here Fx, Fy), and for the requested number of simulations. The analysis flow is used to select the most damaged node in the fatigue simulation. This information can be extended by failure mode, i.e. by component, to track the results of separate hotspots. The DOE results shown in Figure 5 enable the design space to be reduced by retaining only those random variables that are statistically relevant to the results.