simtk.org

Hi all,

I am currently having some difficulties with my pelvis residuals following Inverse Dynamics from the flight phase (no external forces are therefore applied) immediately out of the starting blocks (see attached figure). Whilst the Fy residual is quite high, I am more concerned with the behaviour and magnitude of the Fx residual.

Would anyone who has assessed sprint running (flight phase) be willing to please share their residual forces so I can get an appreciation for the magnitude expected?

I believe I have scaled the model correctly, the IK output seems reasonable (max marker error slightly higher than recommendation - right acromion ~4.7cm) and filtered IK output at 4 Hz prior to performing ID in the GUI. I have attached the Model and IK file if that helps to solve my problem.

Kind regards,

Nicos Haralabidis

Hi Nicos,

I have a few things you can check to see if it helps.

You should check your filtering techniques of your inverse kinematics (i.e. joint angles) and particularly your ground reaction forces. If you're using a standard Butterworth filter with low cut-off frequencies, you're likely to get overshoot in the onset and offset timings of your ground reaction force during sprinting due to the substantial change in force during foot strike and toe-off. Hence, this discrepancy may cause your residuals to be unexpectedly high or abnormal.

I would also check the scaling of your torso and/or your model, especially the mass of your segments making sure that you've included the weight of the subject. Sometimes, high residuals are present because of a lack of balance between your ground reaction force and the calculations of your equations of motion due to mass and inertial discrepancies.

Hope that helps!

Adrian

Hello Adrian,

Thanks for your advice!

I checked through my steps and the odd behaviour of tX seemed to disappear once I filtered the IK output at 8 Hz using the built in OpenSim filter (see Figure 1). I was filtering the data at 4 Hz... likely too low! The data from the first figure is of purely the flight phase, but I still have very strange behaviour of Ty...

Figure 2 and 3 is of the flight phase followed by the contact phase residuals with no GRFs and GRFs applied, respectively. I have not applied filtered GRFs yet as I have noticed the 'overshoot/undershoot' behaviour of the signal during touchdown/takeoff following filtering. I was considering using an adaptive filter as outlined here by Erer (2007) https://ac.els-cdn.com/S002192900700102 ... 4f4f07d131 to try and remedy this particular problem...


I would also check the scaling of your torso and/or your model, especially the mass of your segments making sure that you've included the weight of the subject.

By this do you mean that the combined mass of the scaled model sums to the total mass of the subject? I have checked this and they seem to add up. I perhaps need to consider 'unticking' the Preserve Mass Distribution option whilst scaling, as my subject was more 'top heavy' than usual.

Kind regards,

Nicos Haralabidis

I checked through my steps and the odd behaviour of tX seemed to disappear once I filtered the IK output at 8 Hz using the built in OpenSim filter (see Figure 1). I was filtering the data at 4 Hz... likely too low! The data from the first figure is of purely the flight phase, but I still have very strange behaviour of Ty...

I have a feeling that 8 Hz is still a bit low for the dynamic nature of sprinting. I typically use 15 Hz and I find that a bit suspect. What were you thinking of filtering your GRF with?

Figure 2 and 3 is of the flight phase followed by the contact phase residuals with no GRFs and GRFs applied, respectively. I have not applied filtered GRFs yet as I have noticed the 'overshoot/undershoot' behaviour of the signal during touchdown/takeoff following filtering. I was considering using an adaptive filter as outlined here by Erer (2007) https://ac.els-cdn.com/S002192900700102 ... 4f4f07d131 to try and remedy this particular problem...

I'm not sure if this is the best way to reduce the dynamic inconsistencies that filtering the GRF causes (along with performing RRA), but what I have done in the past for my GRF data is that after I've filtered it, I process two more functions. The first function removes any negative values that appear in the vertical direction as a result of the undershooting effect caused by the filter. Then I use a pre-labelled foot strike and toe-off event identified from the raw GRF data trace to crop the GRF data because anything outside of these events are primarily due to the filtering step, not the period when the foot is in contact with the force plate. However, this will likely cause dramatic changes in magnitudes especially at the instant of foot strike and toe-off in the vertical direction, which is problematic for muscle-driven simulations. I haven't looked into adaptive filters but they might be a better approach.

I would also check the scaling of your torso and/or your model, especially the mass of your segments making sure that you've included the weight of the subject.

By this do you mean that the combined mass of the scaled model sums to the total mass of the subject? I have checked this and they seem to add up. I perhaps need to consider 'unticking' the Preserve Mass Distribution option whilst scaling, as my subject was more 'top heavy' than usual.

That's fine. What I meant was that the mass of the upper torso needs to be accounted for if, for example, you're using a lower-limb model only. If not, you'll end up with large residuals at the pelvis.

Adrian