Fair enough Daniel. I have confidence in the curve reliability so...

By the way, in your first reply you recommended a "cyclic nonlinear analysis for 20 load steps and compare the results to a design criteria". In my case, besides the 20.000 cycles fatigue, I also have to assess the system behaviour (with a different loads setup) during 20 cycles but my first cycle already includes 5 steps which required several hundred iterations to converge so, if I understood correctly your proposal, repeating this over 20 times would take ages... Do you know any alternative and shorter way to evaluate the accumulated impact of these 20 cycles?