Solutions for noisy treadmill force data?

[Maria Fox]

Does anyone have suggestions for how to deal with noisy treadmill force data (specifically X/Z noise)? During walking they look okay but they are considerably noisier during running. I am currently using a 4th order Butterworth filter with cutoff of 10Hz to filter all of my force data, which works well for Y forces, but leaves X and Z forces fairly noisy. I believe this is causing a problem for my ID results as well (although my joint moment results don't look correct, they seemed to be flipped?--this may be a separate issue itself).

I'm attaching a plot of what my L and R forces look like, as well as ID results for R lower limb joint moments and an example file of my GRF data.

Some possible solutions:
- Filter data differently?
- Ignore X/Z forces and only use vertical GRF?
- Analyze time ranges near peak force (i.e. cut time range down to ignore beginning and end of gait cycle)?

Has anyone else dealt with this issue? What solutions worked for you?

[Michael Asmussen]

Hi Maria,

What are you filtering your kinematics at? Was the walking inclined or on even ground?

Also, did you take a look at your center of pressure data? Sometimes when the forces are low, the center of pressure becomes a bit noisy. Check those traces to see if that is the source of your noise.

Best,

Mike

[Maria Fox]

Mike,

Thanks for your suggestions! Both walking and running data are on an instrumented treadmill (no incline). Kinematics get filtered at 8Hz.

My COP data are definitely noisy as well. I will try some different filtering techniques for my COP and moment data and see if that helps anything!

Maria

[Nicos Haralabidis]

Hello Maria,

What do your kinematic data look like prior to and after filtering? As this filtering may be changing the kinematics drastically and causing the spikes in the moments.

Also, have you superimposed your GRF.mot file onto your IK output in the GUI to check how the forces are being applied to the model?

Would it be possible for you to also share a picture of your non-filtered GRF data?

Thanks,

Nicos Haralabidis

[Maria Fox]

Nicos,

These are great points! I checked my kinematic data, and as far as I can tell, the filtering has very little effect because they are already pretty clean. I do not think the kinematic data are causing the problem.

I checked the GRF data with IK results. The GRF vector is placed well under the feet during most of the stance phase during running, but I did notice that the GRF veers way off near/at the end of stance, around toe off. I've attached a couple of pictures of what the GRF looks like around midstance and when you can see it way off the foot. I knew there were some issues with the stability of the COP on the treadmill, so I tried calibrating with an instrumented pole and the procedure did not work because our treadmill has some nonlinear behavior that did not work well with calibration.

Filtering doesn't seem to fix this issue with the COP (it smooths out the noise, but the COP still stabilizes after/before heel strike/toe off). This is why I was wondering if it would be necessary to cut my inverse dynamics analyses to the points of each gait cycle that are stable.

Any thoughts? I've attached a figure of the unfiltered force data as well as the GRF vectors at different points during running. Some heel strikes/toe offs look better than others, but I've attached some problematic examples.

Maria