Large Residuals RRA SO issues.

[Karen Fitzgerald]

Hi everyone,

I am having some difficulties that I'm hoping somebody may be able to help me with.

I wish to use the gait2392 model to perform static optimization to generate muscle forces during normal gait.

I have successfully scaled the model (keeping maker errors below 2cm for max error and RMS less than 1cm).

Inverse Kinematics has also been successful (keeping maker errors below 4cm for max error and RMS less than 2cm) and verified against outputs from motion capture software and previously published literature.

When I move to the static optimization phase I am getting large residuals particularly at heel strike and toe off (note data is for stance phase of gait as only one clean force plate strike available for each trial), thus I must apply strong residual actuators (optimal force 10, control plus minus infinity) for FX,FY,FZ,MX,MY,MZ to get SO to run.

residualForces_SO.png (150.44 KiB) Viewed 901 times

I cannot get these residuals below the recommended 5 percent or less (peak and RMS) of the magnitude of the experimentally measured net external force (50N).

I followed all the troubleshooting tips advised here to no avail (https://simtk-confluence.stanford.edu:8 ... +Practices)

I checked my inverse dynamics results and I get these large residuals for pelvisX, pelvisY,pelvisZ.
This led me to check my GRFs but they seem to be reasonable and applied correctly (visualize GRFs with IK using the preview experimental data function in the GUI)

My next step was to perform RRA.
My problem here is that I cannot get the residuals small enough whilst also following pelvisX,pelvisY, pelvisZ translations i.e. translations less than even 5 cm.

Again I've followed all of the troubleshooting documentation etc but do not know where to go from here.

I've attached my model with all setup files etc if anyone has some time to look at it.

RightFootGait.zip

(370.02 KiB) Downloaded 20 times

Any suggestions or advice is welcome.

Thanks for your time.

Karen

Note of some things I've tried:
-Filtered input data, grf and kinematics
-Checked for sudden jump in accelerations in data frames
-lock the subtalar and mtp joints in the .osim file.
-Check the pelvis COM location in the Actuator files
-visualize GRFs with IK data

[jimmy d]

There are dozens of forum posts with this same question-- I would suggest starting by looking through the forum for what other people have done.

The typical issue is that kinematics and grf's were filtered at different frequencies. Also, depending on your method of calculating COP, there will be errors in the COP at the beginning and end of gait events.

[Karen Fitzgerald]

Hi James, thank you for your reply.

There are dozens of forum posts with this same question

Perhaps I am searching with the wrong keywords but have not been able to find any additional information on the forum that is different to that of the troubleshooting tips provided in the documentation.

The typical issue is that kinematics and grf's were filtered at different frequencies.

By this do you mean the sampling frequency of the data?
Kinematics and GRF data have a sampling rate of 100Hz (i.e. 0.01 s) output from the motion capture software (Codamotion v6.78.1 software).

With regards to filtering, the motion capture software filters the kinematic data with a 4th order 8Hz butterworth filter and the GRF data at with a 4th order 20Hz butterworth filter.
As a check I filtered raw data of both kinematic and grf with a 4th order 8Hz butterworth filter to check (see attached model) but get the same results.

Also, depending on your method of calculating COP, there will be errors in the COP at the beginning and end of gait events.

The motion capture software outputs the COP data relative to the centre of the top surface of the force plate and I just transform this from plate co-ordinate system to model co-ordinate system.

I checked for errors in COP at the start and end of gait and reduced them somewhat but it made little/no difference to the SO results.

The only thing that changes the residuals is modifying the peak Y ground reaction force at heel strike and toe off which invalidates the data.

If you had any other suggestions I'd really appreciate them.

Thanks for your help.

Karen

RightFootGait8Hz.zip

(425.41 KiB) Downloaded 21 times

[Karen Fitzgerald]

Hi all,

So I was able to replicate my results using the kinematic and ground reaction force data provided with the gait2392 model.

Simply by removing the left GRF data, even though I am looking at right leg stance phase, I get these large residuals at heel strike and toe-off. Thus any period of double support is problematic.

Clearly this is a limitation of my data as the gait lab I used only had two force plates, thus I only have a clean strike for a single force plate and a partial/no strike for the second force plate.

Does anyone have any suggestions on the best way to combat this limitation or some best practices in this case?

Thanks again for your help,

Karen