Prior modeling efforts enabled us to simulate the performance of a housing attachment in a handling—drop environment. This project applied verification and validation techniques to quantitatively compare model predictions with experimental measurements in order to evaluate the adequacy of our computational models and modeling techniques. We used the PRESTO analysis code and performed computations on ASC White and ASC Q computing platforms. This V&V project was a milestone in our ongoing research for NNSA’s Advanced Simulation Computing initiative.
Quantitative comparisons between test and analysis results provide a more useful evaluation of the accuracy of a finite element model than the traditional means of side-by-side comparison or overlaying of results. Once the quantified baseline error in comparison is known, uncertainty quantification techniques can be used to determine which parameters most affect the comparison and what the range of uncertainty in the comparison of results actually is, due to the uncertainty of knowing what those parameters should be.
In this study, we used a hierarchical approach with three levels of geometric complexity, from a simple screw to a tab-with-screws subassembly to a system mock subassembly. Experiments provided useful validation-quality measurements to compare with the finite element model results.
At the screw level, we applied an increasing displacement across the screw until the screw failed, and measured displacement at the point of failure. Parameters investigated included load angle and load rate.
Side-by-side comparison showed “reasonable” comparison with test data, whereas a quantified comparison showed that the model was not correctly capturing the measured behavior at all loading angles.
At the tab-with-screws level, we applied an increasing displacement across the subassembly until all of the screws failed. Displacement at the point of first screw failure was measured. Parameters investigated included loading orientation and load rate.
As before, side-by-side comparison showed “reasonable” comparison with test data, whereas a quantified comparison indicated that the model was not correctly capturing the measured behavior at all loading rates. An uncertainty quantification study showed that the uncertainty in the coefficient of friction resulted in a large variation in the predicted results.
At the system mock subassembly level, we applied an impact loading to the subassembly at the housing. We then measured the time to screw failure at each tab. Parameters investigated included loading orientations and drop heights.
Again, side-by-side comparison showed “reasonable” comparison with test data, whereas a quantified comparison showed that the model was not correctly capturing the measured behavior for all tests. At this validation level, test data quality was marginal with a large range of uncertainty. An uncertainty quantification study again showed that uncertainty in the coefficient of friction led to a large variation in the predicted results, as did uncertainty in orientation and order of failure.
This study quantified the potential error associated with the model predictions. The finite element models can be used in conjunction with this quantified error to provide defendable predictive capabilities for scenarios for which no test data exist.