Understanding fatigue and ultimate loading

Some basic background before i ask the question. I have been using openFAST, am opensource software to simulate a 5MW wind turbine. I then use the data from openFAST, and run it through its counter partner Mlife, to calculate fatigue and time to failure.

Below is the data that Mlife dumped to an excel file. time is in seconds, so 6.5*10^8 seconds is 20 years'ish. m is the Woeller exponent. RootFxb1 is the force in x direction at the root of blade 1. RootMxb1 is the moment.


Here is what i am trying to understand. At 12.5 m/s, the ultimate load is 806kN and will last roughly 20 years, but at a MUCH SLOWER wind speed, the ultimate load is drastically reduced to 280kN, but still lasts 20 years. This tells me that at slower wind speeds, the blade is weak and can only handle 280kN. This seems wrong because the ultimate load should be 806kN no matter the speed.

IE, if the blades ultimate load is 806kN, then at 0 wind speed, the blades ultimate load should still be 806kN. So why is the ultimate load so different?

nothing settings wise has changed at all besides changing the wind speed. (not that you guys know the software, just an fyi)

not sure if this helps or not, But for the wind speed of 12.5m/s, i would adjust L_ult for all 4 categories (RootFxb1 etc etc) until the time was 20 years.

thanks for any help understanding the data

Does your software have a curve for frequency of wind vs velocity of wind? There should be a lot more time at 4.5 m/s than at 12.5 m/s, so more fatigue cycles. That, plus the Woeller exponent could explain the results.

It would be interesting to know what the software calls “ultimate load”.
It seems to be the maximum load in the fatigue cycle.

What changes with less wind speed is the pitch of the blades, the bending load and the centrifugal force, which combined with the rotating weight vector of each blade, could increase or decrease the cyclic maximum tension and compresion loads.

A greater difference between wind speed between the higher and lower points respect to the ground, which induces a cyclic bending load, could also change with increased wind speed.

Are you working on the fatigue of the blade’s root or on the fatigue of the pitch bearings?

jrmichler said:

Does your software have a curve for frequency of wind vs velocity of wind? There should be a lot more time at 4.5 m/s than at 12.5 m/s, so more fatigue cycles. That, plus the Woeller exponent could explain the results.

both wind files that i generated were set for 3600 seconds of wind for each respective speed. So both time series data had 60,000 time stamps or close to it.

Here is the only graph i have of the wind:)



thanks

Lnewqban said:

It would be interesting to know what the software calls “ultimate load”.
It seems to be the maximum load in the fatigue cycle.

What changes with less wind speed is the pitch of the blades, the bending load and the centrifugal force, which combined with the rotating weight vector of each blade, could increase or decrease the cyclic maximum tension and compresion loads.

A greater difference between wind speed between the higher and lower points respect to the ground, which induces a cyclic bending load, could also change with increased wind speed.

Are you working on the fatigue of the blade’s root or on the fatigue of the pitch bearings?

the theory manual defines L_ult or ultimate load is as follows:

where L_ult is the ultimate design load of the component.

thats all the information i can find. If it means the ultimate stress over a cross section, then when i try that number, my life time is infinite since it spits out 726,000,000 years as the result. So i personally think the L_ult is the max loading that occurs in the time series data.

What i did was kept adjusting the L_ult in the software until it spit out a time of 20 years. So for a wind of 12.5m/s and a MAX loading of 444kN on the blade during the time series data, it says the L_ult is 806kN to get a lifespan of 20 years.

But if i test out the data for a 4.5m/s with a MAX loading of 180kN, it spits out a L_ult of 280 for a life span of 20 years. This seems way way off. At slower speeds, those blades should be able to hanlde more than 280kN

I have the software setup to look at the root of the blade where the threaded m45 studs are. Here is a picture:

What is your software really looking at?

The strength of the studs?
The threaded fasteners on the studs?
The attachment of the studs to the blade?
The blade itself away from the studs?
The stress concentration region between the stud attachment and the blade "far" (St Venant's principle) away?
Something else?

What are your inputs to the software?

Are you defining an S/N curve?
A stress as a function of blade loading?
The location of the blade loading - distance from the blade root? Are you examining the mean and alternating stress in the output?
Do those outputs match hand calculations for the combined centrifugal and wind loading blade forces?
Have you loaded some extreme cases into the software, such as spinning with zero wind load and wind load at zero speed and checked the results against hand calculations?
Etc.

When using a new software package, it is always necessary to find exactly what that software is telling you. You do this by checking limiting cases against hand calculations, experimental data, and other proven software. In your case, the software is probably giving correct answers for the inputs it was given. But you do not know that, so you need to test the software. You might find that you are incorrectly using the software. Or you might find a bug in the software.

Back when I was using a well known software, Abaqus, to model sheet metal bending, I used the same element that other sheet metal researchers had used. When my FEA results were 30% different from my experimental results, a sheet metal research professor told me that it was experimental error. He was wrong. It was an element formulation that correctly modeled tension, but not bending.

Keep at it until you fully understand the results, even if it takes a few weeks.

jrmichler said:

What is your software really looking at?

The strength of the studs?
The threaded fasteners on the studs?
The attachment of the studs to the blade?
The blade itself away from the studs?
The stress concentration region between the stud attachment and the blade "far" (St Venant's principle) away?
Something else?

What are your inputs to the software?

Are you defining an S/N curve?
A stress as a function of blade loading?
The location of the blade loading - distance from the blade root? Are you examining the mean and alternating stress in the output?
Do those outputs match hand calculations for the combined centrifugal and wind loading blade forces?
Have you loaded some extreme cases into the software, such as spinning with zero wind load and wind load at zero speed and checked the results against hand calculations?
Etc.

When using a new software package, it is always necessary to find exactly what that software is telling you. You do this by checking limiting cases against hand calculations, experimental data, and other proven software. In your case, the software is probably giving correct answers for the inputs it was given. But you do not know that, so you need to test the software. You might find that you are incorrectly using the software. Or you might find a bug in the software.

Back when I was using a well known software, Abaqus, to model sheet metal bending, I used the same element that other sheet metal researchers had used. When my FEA results were 30% different from my experimental results, a sheet metal research professor told me that it was experimental error. He was wrong. It was an element formulation that correctly modeled tension, but not bending.

Keep at it until you fully understand the results, even if it takes a few weeks.

thats what is difficult with this software. I do not define really much of anything and it just does the calculations. there is A LOT that runs in the background that i am unaware of and do not have access to.

None of the files i have access too, state what openfast is looking at or doing for its calculations. Mostly nodes from Aerodyn and elastodyn.

All i have defined is the wind speed, and then the 100 scripts, then calculated 52 different channels of data. I do not define a distance or a load or anything. It calculates the loads and moments based off of rainflow counting and wind speed.

One of those is called the RootFxb1 which is the force in x direction of the root of blade 1. Root is define as the end of the blade that hooks to the hub. As far as i know, studs are not part of the equations.

Here is an image of what openfast calculates based on the wind file of 12.5m/s. The left hand side are are the different output channels and their associated graphs.

All i have to do is edit a basic text file and have it point to the wind file, then openfast does the rest.

Mlife then takes the data generated from openfast and computes the fatigue based off of the channels i tell it to use. I was informed to use RootFxb1, RootFyb1, RootMxb1, and RootMyb1 as the channels for it to use.

i tried to check against hand calculations, but the issue is, the output does not give me enough data to use the formulas. It gives me about 1 or 2 values, which then leaves me with 2 to 3 unknowns. And since i do not provide ANY inputs, it makes it very difficult.

thanks :)

Sure looks like the best bet is to write-off this software package and get one that actually Works.

At present, it is a losing proposition.

Tom.G said:

Sure looks like the best bet is to write-off this software package and get one that actually Works.

At present, it is a losing proposition.

What would you suggest?Thanks

Wish I could be of some real help, unfortunately that is WAY out of my field (mostly Electronics). My comment was based on the almost total opacity of the program you are using along with its nonsense results.

@jrmichler is one of the resident mechanical experts here, and mentioning him like this will send him a notification.

Tom.G said:

Wish I could be of some real help, unfortunately that is WAY out of my field (mostly Electronics). My comment was based on the almost total opacity of the program you are using along with its nonsense results.

@jrmichler is one of the resident mechanical experts here, and mentioning him like this will send him a notification.

I highly appreciate the help and willingness to find a better solution. I agree that its been a pain the last few weeks.

The OpenFAST software seems to be a solid package. That being the case, it is clear from the documentation that it has a significant learning curve. I suggest diving into the Testing OpenFAST section of the user documentation: https://openfast.readthedocs.io/en/main/source/testing/index.html. Play around in there until you fully understand how everything works. Look very carefully at all assumptions in the software, especially for anything that could cause increased stress at lower wind speeds. Such as the effect of blade pitch on peak stress at some particular point.

LT72884 said:

. At 12.5 m/s, the ultimate load is 806kN and will last roughly 20 years, but at a MUCH SLOWER wind speed, the ultimate load is drastically reduced to 280kN, but still lasts 20 years.

Since a 5 MW wind turbine is a $10,000,000 machine, and you are clearly working by yourself, can we assume that you are a student working on a student project? This sort of checking is a sign of a good future engineer. Do you mean "ultimate load" or "peak load"?

jrmichler said:

The OpenFAST software seems to be a solid package. That being the case, it is clear from the documentation that it has a significant learning curve. I suggest diving into the Testing OpenFAST section of the user documentation: https://openfast.readthedocs.io/en/main/source/testing/index.html. Play around in there until you fully understand how everything works. Look very carefully at all assumptions in the software, especially for anything that could cause increased stress at lower wind speeds. Such as the effect of blade pitch on peak stress at some particular point.

Since a 5 MW wind turbine is a $10,000,000 machine, and you are clearly working by yourself, can we assume that you are a student working on a student project? This sort of checking is a sign of a good future engineer. Do you mean "ultimate load" or "peak load"?

Yes I am a student haha.

The user manual for openfast and mlife say "ultimate load"

The dev of OPENFAST told me via the GitHub forums that ultimate load can be found 2 ways.

First is FEA
And second is by taking the max load of the time series data, and scaling it by a factor, until your life time hits 20 years. This is the method I went with since I do not have fea.

I just find it odd that at such low wind speeds the L_ult is 280 while L_ult of the faster windspeed is so much higher.

I have double checked all settings in openfast and all are the same except the windspeed.

Since this is a student project, I moved this thread to the Engineering Homework forum.

The fact that you changed only one input does not mean that nothing else changed. When working with a complex software package, keep in mind that the software might be changing internal variables. Some possibilities include:

Blade pitch angle, which will affect stresses
Number of load cycles
RPM

What controls the blade load? It could be the maximum torque of the generator, the RPM, or something else. Look for things that change at high and low speeds. For example, if the blade pitch is changed because of a generator torque limit, then the direction of the wind force vector relative to the blade chord will also change. That will have a significant effect on blade stresses.

It is your job to find out what the software is really doing. Just feeding it inputs and getting results is not enough. Spend some time studying Post #6. When I was learning Abaqus, I spent well over a hundred hours doing that sort of investigation of the software before attempting to run a real problem.

jrmichler said:

Since this is a student project, I moved this thread to the Engineering Homework forum.

The fact that you changed only one input does not mean that nothing else changed. When working with a complex software package, keep in mind that the software might be changing internal variables. Some possibilities include:

Blade pitch angle, which will affect stresses
Number of load cycles
RPM

What controls the blade load? It could be the maximum torque of the generator, the RPM, or something else. Look for things that change at high and low speeds. For example, if the blade pitch is changed because of a generator torque limit, then the direction of the wind force vector relative to the blade chord will also change. That will have a significant effect on blade stresses.

It is your job to find out what the software is really doing. Just feeding it inputs and getting results is not enough. Spend some time studying Post #6. When I was learning Abaqus, I spent well over a hundred hours doing that sort of investigation of the software before attempting to run a real problem.

i have about 60 hours into it so far.

Since i cant see the coding, i cant see what happens when i change one single variable like the wind speed. I guess i can ask the developer and see what he says.

I have read a lot of the user documentation that NREL wrote, but it doesnt always explain everything or in the best way haha. However, i am still investigating what is going on.

I have brought this to his attention that its strange at slower wind speeds, the ultimate load that the blade can handle is drastically reduced. The dev did say that changing the wind speed should not have done that.

Ill go back and read post 6 as well as the link you posted.

thanks again for the advice.

The sort of problem that you are experiencing casts doubt on all of the software outputs. You need to keep working until everything makes sense.

LT72884 said:

I have brought this to his attention that its strange at slower wind speeds, the ultimate load that the blade can handle is drastically reduced.

You might consider submitting a bug report. I once had a similar problem when using an academic copy of Abaqus. Academic copies did not come with technical support, so I submitted a bug report. They looked into it, and responded that I had found a bug.

Another possibility is to search wind turbine blade failures for blades that failed due to fatigue, then analyze those blades with the software to find if the software correctly predicts the failure.

jrmichler said:

The sort of problem that you are experiencing casts doubt on all of the software outputs. You need to keep working until everything makes sense.

You might consider submitting a bug report. I once had a similar problem when using an academic copy of Abaqus. Academic copies did not come with technical support, so I submitted a bug report. They looked into it, and responded that I had found a bug.

Another possibility is to search wind turbine blade failures for blades that failed due to fatigue, then analyze those blades with the software to find if the software correctly predicts the failure.

im going to try and submit a bug report. Reason being, after reading multiple studies and research papers over the course of the last 8 or 9 days, my results for the first simulation fall within the same range as the papers. However, at the slower wind speeds, the papers ultimate load didnt change due to wind speed.

ill double and tripple check everything before reporting the error.

thanks