Simulating Landing Accelerations in NASTRAN for Fatigue

[vorwald]

I have carrier landing acceleation data of a pod, and have created a NASTRAN model of the pod. I would like to
a) represent the landing accel data in NASTRAN for fatigue analysis
b) evaluate the landing accel data vs airspeed to identify trends for extending to other landing speeds.

So far, the best approach I've come up with is to integrate the accels twice, pass the estimated displacement through a high pass filter (20 Hz) to remove rigid body effects, and implement the remaining "elastic" displacement in NASTRAN using spcd cards. Minimum structural mode appears to be 26 Hz.

This "static deflection approach" ignores dynamic / modal stress distribution. Also, the plots of peak deflection vs airspeed has no meaingful trend on linear/semilog/loglog plots.

Right now, I am investigating an alternate approach. Basically, I'm specualating that the measured accel data consists of structual modes excited by impact forcing. So, I passed the accel data through low pass filtered the accel data to remove high freq content (500 Hz), and subtracted of the mean for the first .1 sec. I'm using an analytical response of the modal equation driven by dirac forcing, which has been programmed in a form for multiple modes; each mode has a natural frequency and damping, and each mode can have multiple impulse loads with individual amplitudes. I am working through the acceleration time history, trying to specify impulsive loads (dirac time, amplitude), frequency, damping, and mode shape (at the gauges) to match the measured acceleration data. Basically, I'm guided by the time period of the next one to oscillatory cycles of the primary gauge(s), at that time period. I hope that modal impulsive forcing identified via this approach will have a more meaningful trend with landing speed.

I haven't thought about how to transfer the frequency; damping; modal forcing to NASTRAN. Could fix the damping via table to allow select modes to response, and apply forcing at dominate gauge for that time period. One thing, I not sure how to seperate out response that would be due to airframe flexure; vs pod flexure.

Any suggestions.

 

[40818]

vorwald,
You post has left me somewhat confused/overloaded with info.
Your post obvioulsy made clear sense to you on writing it, but it is kinda hard to follow what your asking/trying to achieve??

 

[rb1957]

I guessing he's got a mountain of data (every 1/100th of a second maybe, and trying to make sense of it.

I'd look at each degree of freedom separately (at first, you'll see). I imagine you've got accel's for all 6 degrees of freedom. pick out the extreme values of each, this'll be the largest cycle. pick out the next extreme peaks; maybe you can rationalise a value that represents several cycles. if the load range is 1/2 of the previous cycle then the damage per cycle is about 1/16. you've also got an endurance limit that allows you to say very small stress cycles have a negligible effect on the fatigue life.

recognise too that this is just one phase of the flight, maybe the only phase that you are presonally interested in. I think an important part of the problem is to gauge how severe is the impact compared with the flight, then use this to guide you in how much detail to go into.

good luck !
(I once had to do the fatigue analysis of a sea harrier going off a STOL ramp, so i think i know a little of the mountain of data you're looking at !)

 

[vorwald]

I'll try to restate the procedure above. You can solve the basic m xddt + c xdt k x = F*dirac(t-t0), for a set of modes, like five or 7. Here, x, is the modal dof. So, for each increment in time, try to guess the frequency from looking at the landing data, which appears to consist of several impacts. Then adjust F, and the mode shape at each gauge, to give you the reponse over the next short period of data (like 0.01 sec). When the anaytical response differs from the data, either add a new impulse to one of the modes that was identified up to that point, or add a new mode, primarily based on the frequency content.

So, in MatLab, I'm trying to solve y = sum(phi(j)*x(j)), xddt + 2 * zeta * omega * xdt + omeag^2 * x = sum(F(i)*dirac(t-t0(i))), by guessing omega, zeta, and tunning F(i), t0(i), and adjusting phi(j) to match the dominate gauges at that moment. So, I was slowing making my way though the landing impact response. I like this approach because it makes physical sense to me. But, I don't know how to automate determing when a new mode is needed, or zeta/omega estimates.

Now, I'm on a different, easier path. Use SPCD in NASTRAN to specify the measured accleration at the node/dir in the FEM. So far, I've plotted the NASTRAN acceleraions, and they match the filtered data very well, and the actual data pretty well, up to the cutoff frequency.

With the NASTRAN model loaded to regenerate the acceleration data, getting fatigue damage is reasonable straight forward. I'll just use rainflow, with S/N fit.

The data is at 10 KHz, and there is significat response at 1.8 KHz.