PSI - Issue 19

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

165

16

Table 3. Reliability statistics

= + 553,386 1.555 447,917 148,610 596,527 2.030 417,579 156,400 573,979 4.240 553,386

Test/simulation Full simulation

Material variability Error on FEA stress 1.989 126,357 390,550 516,907 A comparison between the test and simulation results show good correlation. Fig. 12 shows that the simulation results lie mostly within the 90% confidence bounds of the test results, except at large values of life. Analysis from Fig. 13 confirm that probability (Test < Simulation)=0.514. This implies there is almost equal probability that a test life may be lower or greater than a simulation prediction. Table 3. confirms the correlation of the values of the location parameter θ from test and simulation. In this case, θ is used for comparison purposes instead of η because test data are analyzed using a 2-parameter curve whereas simulation data are analyzed using a 3-parameter curve. The first row of Table 3. gives the curve fitting parameters for the measured test data. The second row gives comparable values based on a full simulation. The full simulation considers the statistical variability associated with both material scatter and FEA modelling errors applied simultaneously. The third row considers the effect of material variability alone – the FEA errors are not considered in this case. Similarly, the fourth row considers only the FEA errors. These last two rows are useful why trying to determine whether the variability of one parameter is more or less significant than the other. This is more clearly seen when considering the plots described later. The values of β>1 given in Table 3. are indicative of a fatigue-related failure which is expected in this case study. However, the apparent value of β=1.555 for the simulation compared with β=4.240 for the test requires some explanation. It is inappropriate to compare β values directly for 2 and 3-parameter Weibull analyses. It is also necessary to consider the effect of a failure-free life γ. A comparison of the Reliability vs. Time plot is shown in Fig. 14. The median values presented on the plot show good correlation. The simulation data lie within the confidence bounds of the test data at the 90% level.

 Test data - - 90% confidence bounds  Simulation

Fig. 14. Weibull Reliability vs. Time plot comparing test and simulation data