Modal testing as the most efficient way to verify FEM accuracy? 1

[Mark172]

Hi All,

I lead a mechanical team at a growing aerospace company building small satellites up to 500 kg. I've been tasked with bringing structural analysis in-house to rely less on outside contractors. I intend to accomplish this through a mix of training and hiring, but I am concerned about the learning curve required to build accurate but efficient FEA models with such a young team - only I have significant structural analysis experience (~5 yrs) and even so I'm not really a SME.

I know there is some art to FEM building, and training will take time. My nightmare scenario is a poorly built FEA model that gives spurious modes, stresses, and/or bolted joint margins, which we don't discover until the spacecraft is on a vibe table. Best case scenario we would then have to scramble to recompute the notching plan while we burn money at the test facility. Worst case would be damage to the spacecraft and that is simply not an option.

So I want to be able to confidently vet our FEA models. Having experienced outside analysts review our models is expensive long term and only goes so far. From what I've seen, even the most experienced teams of analysts are routinely off on first modes by ~10%. Without observational data there are always uncertainties. So in my mind the only reasonable means to verify FEM accuracy is testing. We strongly prefer to be able to test in-house, but we don't have the capital for a vibe table of sufficient size. My sights have been homing in on modal testing as the most cost- and time-efficient way to verify our FEMs. My thinking is to purchase a large granite table as the fixture, acquire modal testing equipment, train our analysts through an iterative FEM building - testing - correlating process on subassemblies as we build our next few spacecraft, establishing the best practices for bolted joints and other modeling aspects particular to our buses. Then we would perform additional modal testing on the integrated spacecraft, correlating the vehicle-level model and confirming our notching plan long before the vehicle is at the test facility.

Is this sensible thinking? Am I being overly skeptical of FEA models and will this be a waste of resources? Is there some other type of testing or approach to model verification that would be more appropriate?

TL;DR - I'm trying to build a structural analysis team from scratch and want to make sure our FEA models are accurate. What is the best way to do this?

 

[rb1957]

how accurate are your current (out-sourced) FEMs ?

Do you need your own vibration testing lab ? can you out-source ? (to the guys who test your satellites now ?) or is this part of doing everything yourselves (being the "masters" of your own destiny ?

Car people (like Greg) do a lot in this field.

I'd spend some time understanding the state of the art, and where people who do this now think this field is going.

GL !

"Hoffen wir mal, dass alles gut geht !"
General Paulus, Nov 1942, outside Stalingrad after the launch of Operation Uranus.

 

[Mark172]

Our current outsourced FEMs are sufficiently accurate. They vary by approx. +/- 7-10% for first modes. I'd like to produce in-house models that nail all the primary and most of the secondary modes to within just a few percent.

We don't have the capital for our own vibration testing lab at this time. We could outsource sine sweep testing, but that is expensive long term, imposes transit and handling risks to the hardware, and eats up schedule. We are very schedule sensitive. Modal testing equipment is within our budget and I'm confident our existing team can be quickly trained on it.

 

[SWComposites]

you are very right to be cautious of FEMs. There are a LOT of ways to screw them up.

training needs to start with simple models of simple structures that can be compared to test data. then add in a bit of complexity one at a time.

modal testing will give data to validate model stiffness and vibration frequencies, which is the relatively easy part to predict with FEMs. modal testing will tell you nothing about strength and thus won't be useful to validate FEM based margins, failure modes, etc.

do you have enough previous program test data that could be used to compare to internally created FEMs to use for training purposes?

 

[GregLocock]

You are over complicating what you need to do modal analysis. At its simplest you need a load cell, a hammer, an accelerometer, and some bungee cords. Then it dangles free free and you spend many happy hours bashing the satellite with the hammer.

That first hit with a hammer will give your FEA dudes enough data to keep them busy for months. Good quality hammer testing is a learned skill.

Slightly more elegant and repeatable is to use a voice coil type shaker, and triaxial accelerometers. Your FEA model may need to include the mass of the accelerometers. I suggest if you can to use the shaker dangling from the structure, but it can be floor mounted. Typically we use bandpassed white noise excitation, but swept sine gives the nicest data.

One big thing you get from modal is the damping of each mode. That is crucial for your fatigue predictions.

Of the modern systems I have only used LMS, which has a perfectly acceptable modal analysis suite but is overkill for your needs (4 channels is where I'd start). Prosig probably can supply a turn key system (advert) that will be more reasonably priced. QC on the input data is the be all and end all of modal analysis. It is boring, subjective, and vital, and is where cheap PC based systems can be problematic. Alway check linearity and (easy with a hammer) reciprocity. For hammer testing you need some signal processing malarkey that shaker testing doesn't need.

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[Mark172]

Hi all, thanks for your comments, I've spent time reading up on your pointers. Very useful.

@SWComposites, you say that modal testing tells you nothing about strength, can you explain that a bit? My thinking is that if you've got the stiffness correct, as exhibited by matching the first 5 modes or so, then you're probably predicting local stresses in your structures fairly well. Bolted joints are assessed in a separate analysis using extracted CBUSH forces at the joints.

@Greg, I appreciate the nudge toward keeping it simple. It seems to me there is benefit in mirroring the boundary conditions we see on the launch vehicle, that is, fixed-free at the mating interface. If I did go fixed-free with a granite table, are there rules of thumb for the fixturing to make sure measured modes aren't confounded with fixture interactions? For random vibe testing I know the rule of thumb is staying an octave higher than your test article (so if testing out to 300 Hz you'd need a fixture with a first mode of >600 Hz). Is there a rule of thumb for mass? Say, 10x the mass of your test article?

 

[SWComposites]

Modal tests do not (usually) result in structural failure, so they do not provide any strength and failure mode data for which to correlate to analysis.

if you've got the stiffness correct, as exhibited by matching the first 5 modes or so, then you're probably predicting local stresses in your structures fairly well nope, stiffness is a more or less global response, strength is controlled by local stress concentrations, etc. Your model can accurately predict modes and global stiffness but could be completely wrong in terms of local stress values. But it somewhat depends on how you are doing the stress analysis. If your FEMs are just used to get load distributions in members (e.g., using a beam model for the structure for modal analysis) and you use detailed hand calcs with the FEM loads as inputs to calculate margins, then the FEM may be ok to be validated with a modal test. However, if you have a detailed FEM (shells, solids) where you are extracting local stresses to calculate margins, then you need a much more complex validation path, likely needing full vibration or static load tests to failure.

 

[GregLocock]

On that, our NVH FEA models use elements about 10 times as big as our fatigue ones do.

I think you'll end up modelling both the rig and the satellite, to account for interactions. What do you mean by fixed-free?

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[GregLocock]

Incidentally, you don't need to do base excitation, you can plonk whatever seismic mass you have on the ground, or even bolt whatever clamping mechanism you are using direct to the floor, and then suspend the shaker from a gantry to shake the satellite itself. You are not trying to replicate a real life loading condition, you are validating your FEA. I found this , guessing you'd be using the large ring type system on P9 with pyro bolts?

http://www.eurockot.com/wp-content/uploads/2012/10/Ch4UsersGuideIss5Rev0.pdf

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[TG_eng]

Very interesting conversation here. I don't want to get off topic from the original post but I saw comment from @GregLocock and I'm just curious for learning purposes. @GregLocock could you elaborate on this?

"One big thing you get from modal is the damping of each mode. That is crucial for your fatigue predictions."

 

[GregLocock]

FEA by itself can't predict damping in any believable fashion.

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?