Uncertainty in finite element analysis of a submarine`s hull
More Info
expand_more
Abstract
The Finite Element Method possess the intrinsic characteristic that, due to the assumptions made for the parameters of the problem being modelled, a deterministic uncertainty on the obtained results is always present. This research makes use of basic statistical concepts to obtain reference values for a possible quantification of this uncertainty. The problem used as basis for the analysis is the determination of the hydrostatic pressure which leads to the failure of the central compartment of a Manatee class submarine, by means of non-linear analysis with imperfections included. To generate a set of possible result outcomes, key parameters used in the construction of a model are varied in order to ensure the production of both model-form and solution approximation errors. The parameters varied are four, namely (1) the in-plane integration scheme, (2) the order of the shell elements (3) the small displacements assumption and (4) the mesh size. The produced variation is then grouped in sets related to three different possible analyst’s levels of expertise and the uncertainty is quantified for the data both within and in between groups. The analysis show that only a small uncertainty can be expected in the results from non-linear models constructed by Expert Performers (with the 95th quantile being up to 3.94% larger than the estimated average), but larger scatter is present if outcomes from Advanced Beginners are considered (the maximum 95th quantile in non-linear results is 10.30% larger than the estimated average). If no information is provided on the expertise level of the analyst who produced a model, from comparing the results in between sets, an uncertainty of up to around 12% might be expected for the pressure which induces yield. Other results of interest are described throughout the document, like a post flange yield result which can put current analytical assumptions of global collapse into a different perspective and the impact that varying the parameters have on the first yield location.