PSI - Issue 19

A. Halfpenny et al. / Procedia Structural Integrity 19 (2019) 150–167 Author name / Structural Integrity Procedia 00 (2019) 000–000

166

17

Fig. 15 shows a comparison of the β and θ contour plots at the 90% confidence level. Contours show test data (black), full simulation (red) and the uncertain parameters for material (blue) and FEA stresses (green). It is clear from the size and shape of the contours that material uncertainty contributes more to the overall simulation variability than the error arising from FEA stress analysis. This would suggest that improvements to the FEA accuracy are unnecessary. Material scatter, an aleatoric uncertainty, is the controlling factor in this case. As all contour regions overlap, this implies all analysis cases are not statistically different from one another or from the test results at the 90% confidence level. If the confidence level were reduced, then the contours would also shrink and at some level would no longer overlap. In this case study, a critical confidence level between the test and full simulation is found at the 82% level.

Fig. 15. Weibull contour plot comparing test and simulation with uncertainty in material and FEA stresses

In conclusion, the simulation is found to give good correlation with the experimental results. The simulation should offer a reliable platform on which to perform additional design simulations and for extrapolation of the reliability statistics. However, it should be remembered that only one pressure level was used in the tests and further testing over a range of different levels would be advisable if the simulation were to be used for variable amplitude loads. 5. Conclusion This paper has demonstrated the advantages of using probabilistic fatigue simulation and stochastic design over the traditional deterministic design approach. It has demonstrated how Monte Carlo simulation with Latin hypercube sampling are effective for obtaining simulated reliability tests based on standard FEA models. It also described how design robustness is ensured by using factorial sampling in conjunction with a response surface model. The requirements for a Reduced Order Model, obtainable from FEA, were discussed. A review on reliability analysis based on simulated fatigue analysis data was presented. Particular attention was paid to recommendations for addressing the effects of fatigue endurance limits, static failures and non-linearities in