FTA Models best practice

medini_1050701002

Hello everyone, With the team we argued about what was the best practice when doing FTAs. Our situation is the following: - We have several system level fault trees - Each leaf in the system fault trees is detailed in a subsystem fault tree  - We also have copies of the system fault trees in which each leaf is replaced with a transfer gate to the corresponding subsytem fault tree top event.  Until now we had one Medini "FTA model" per tree (that would equate, in our projet, to 100+ FTA models in the system medini project).  Is this considered a valid approach our should we have, for example, one FTA model for the system containing all the system FT and one FTA model for each subsystem? It doesn't seem to matter in terms of functionality (we're able to do everything we want this way) but could it change something performance-wise?  Adrien Gérard

medini_1023909254

Hi Gerard, there are pro's and cons between both approaches: In case you have one  FTA-Model containing multiple fault trees (and diagrams) you can make efficient use of MOE's derived from architecture elements/failure modes. The Outline view would always give you a quick view on all the MOEs and also on how often they occur in your FTA model (which may be indicators for common causes). If you would use instead individual FTA models for all the trees, than you would get a lot of transfer gates modeling explictly the multiple occurring events. In that case you would have to investigate manually, how many transfer gates would point to the same target. (Note: The detection of MOEs and the information on how many times they do occurr is always scoped by an FTA model) On the other side, if you want to share  sub-trees between many different higher level trees, it would usually be better to have that tree as a separate model in order to avoid copying sub-trees between different models (which especially in case of changes always has the risk of inconsistencies). For the calculation of course it does not make any difference whether you directly use MOEs or model them with transfer gates. Best regards, Eckhardt