FAST.Farm and Impact of Seed Numbers on Fatigue Damage

Hi everyone,

I’m running simulations in FAST.Farm to calculate fatigue damage and I am conducting a convergence analysis to determine how many simulations are needed for reliable results.

So far, I’ve run simulations with 30 different random seed numbers, for different wind directions. However, I’ve noticed that a few simulations produce fatigue damage results that are significantly different from the others. For example, in the attached image, you can see that seeds 23, 24, and 25 yield values that deviate noticeably from the rest.

The seed numbers were chosen randomly by me, and I’m wondering if this variability could be related to the specific seed numbers I selected. Is there a recommended method for choosing seed numbers to ensure consistent and representative results?

Intuitively, I wouldn’t expect the seed numbers to have such a significant impact, but since I’m unsure, I wanted to reach out to the forum for insights. Has anyone else encountered similar behavior, or could anyone provide guidance on whether seed selection plays a role in convergence for fatigue damage analysis?

Below are the seed values (seed1 and seed2) I used for these simulations:

seed_dict1 = {
1: (2073448544, -75384093),
2: (1532865329, 65454852),
3: (456789, -5054),
4: (34, 456),
5: (-65453, -5),
6: (-1, -2345),
7: (456345, 344),
8: (6456347, 445654),
9: (2345, 6674),
10: (23452456, 78565),
11: (-456345656, -4534546),
12: (2357, -45634563),
13: (987689, 787666),
14: (-98756545, 6756),
15: (-1234545, -45636574),
16: (-118803747, 247504697),
17: (-171161675, 206919021),
18: (-184742058, 904353589),
19: (-295269453, 555203102),
20: (-301652503, 634537670),
21: (-540227741, 58744843),
22: (-592844958, 878708298),
23: (-677123190, 662101981),
24: (992948224, 95641675),
25: (504323988, -466215084),
26: (717061322, -259797599),
27: (-944537124, 174755444),
28: (268945319, -336690441),
29: (-838449676, 902207459),
30: (925578245, 854775900)
}

Thank you in advance for your help!

Best regards,
Tiago Lucas

Dear @Tiago.Lucas,

Presumably you are using TurbSim to generate inflow for FAST.Farm using Mod_AmbWind = 2 or 3 and the seeds you are referring to are used as RandSeed1 and RandSeed2 in TurbSim; is that correct?

I would normally expect that you’d select the seeds randomly (or sequentially) and you’d assess convergence of some quantity of interest (in your case, tower fatigue damage) by running multiple seeds, averaging across seed, and tracking the convergence of this average. The IEC 61400 design standard calls for a low number of seeds (6 in most load cases), but this is often not enough for a converged solution, which may take 20-50 seeds (depending on the situation) in reality.

I’m not sure physically what is driving large differences in tower loads between the seeds you are using, but perhaps you could analyze the results of each case to identify the physical reason why. Regardless, assessing convergence by running more seeds is prudent.

Best regards

Hi @Jason.Jonkman,

Yes, I am using TurbSim to generate inflow for FAST.Farm with ‘Mod_AmbWind = 2’, and the seeds I referred to are ‘RandSeed1’ and ‘RandSeed2’ in TurbSim, yes.

I’ll analyse the individual cases to better understand the physical factors driving the observed differences in tower loads. Increasing the number of seeds beyond 30 also seems like a prudent step to improve convergence.

Thank you for the guidance.

Best regards,
Tiago Lucas