Announcement

Collapse
No announcement yet.

Filtering kinematics and kinetics for inverse dynamic parameters- what is best prac?

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Filtering kinematics and kinetics for inverse dynamic parameters- what is best prac?

    Hi,

    I am starting another thread to discuss this issue which I have encountered in the BJSM paper by Hewett's group and van den Bogert's group. I am undecided on several things:

    1) Should kinematics and kinetics be filtered at same frequency? Pros and Cons?
    2) If (1) is true, what frequency to filter at? I know kinematics should be based on some sort of residual analysis.
    3) Should the decision of (1) and (2) be based on the type (walk, run, cut, sprint, land, hop), intensity of motor activity, or some other factors? If some please help me in stating these factors...

    References:
    1) Kristianslund et al Effect of low pass filtering on joint moments from inverse dynamics: implications for injury prevention
    2)
    Roewer BD The 'impact' of force filtering cut-off frequency on the peak knee abduction moment during landing: artefact or 'artifiction'?
    3)
    Bezodis NE1, Salo AI, Trewartha G.Excessive fluctuations in knee joint moments during early stance in sprinting are caused by digital filteringprocedures.
    4)
    McCaw ST Filtering ground reaction force data affects the calculation and interpretation of joint kinetics and energetics during drop landings.


    Please throw all opinions at me.

    Regards,
    Bernard
    Last edited by Bernard; December 9th, 2014, 07:44 AM.

  • #2
    Re: Should kinematics and kinetics be filtered at same frequency?

    Probably not - I suspect that if you ask yourself why you are filtering the data, and look at the data itself, then the answer will be fairly straightforward.

    The first question that you need to think about is, "What's the sample rate?" If you are sampling both point and force plate data sets at 50-60Hz then it probably doesn't matter what frequency you filter at because it's not going to change much. On the other hand, if your point data is sampled at 60Hz and the force plate data is sampled at 1200Hz then you need to think about what you are looking for in the data.

    Generally the reason most people give for filtering data is to "clean it up" - you filter point data to smooth the trajectories so that it becomes easier to analyze, and you filter the force plate data to get rid of high frequency noise, impact transients, and sometimes mechanical ringing in the plate. If you look at the data that you are collecting then you should be able to figure out what you need to do - the chances are that you'll filter both streams of data at different frequencies.

    Comment


    • #3
      Re: Should kinematics and kinetics be filtered at same frequency?

      Typically, as stated by Ed Cramp, the kinematics is more in need of filtering than the force plate but for inverse dynamics it is important to filter the kinematics and force plate both with the same filter. Even if the force plate data is perfectly clean. Otherwise, artifacts are introduced in the joint moments during impacts. It's easy to prove that with simulated data where the correct answer is known, such as here.

      This study showed that it's somewhat different for estimating intersegmental forces. At least in running impact where the flexion angles are small and axial accelerations of the body segments during impact are highly coupled. You get better results when force plate data is filtered with a higher frequency than the kinematics (but note that filtering was still needed). For joint moment estimation, however, it's was clear that the exact same filter must be used. Someone should repeat that study with a 3D model and other movements, to have a more robust generalization of those findings.

      I am not a fan of residual analysis, though I admit I have no experience with it. For inverse dynamics of sports movements with impact, the filter frequency can be critical, because the movements and forces have high-frequency components and oversmoothing is a real risk. I generally run the analysis with a sequence of filter settings, and then choose one that still shows some random noise in the moments and forces, but not so much that the signal is obscured. Random noise are high-frequency signal features that are not reproducible between trials. That way, I'm sure I have not smoothed more than I absolutely need to. We always report the filter frequency in our papers. The optimal filter depends on the camera system, sampling rate, and movement, so you can't give a general rule, you have to experiment as I described above. For sports movements with impact, something between 12 and 18 Hz is usually best. The results won't be perfectly smooth, but the remaining random noise can then be absorbed by averaging of multiple trials or by a statistical model. That's better than having smooth results where some real impact load has been removed by the filter.

      More modern approaches to dynamic analysis, such as computed muscle control (CMC) and tracking optimizations, have no need for low-pass filters. If the unfiltered data is tracked by a dynamic simulation model, the system dynamics itself acts as the low pass filter. The filter tuning process is replaced by tuning the feedback gains (in CMC) or tuning the tracking vs. effort weighting in a cost function for optimization. These tunings are easier because results are (in my experience) not as sensitive to the tuning as in inverse dynamic analysis. One thing to watch out for, though, is that these modern approaches require the full body dynamics to be modeled, it cannot be done with a limb or partial limb. If upper body dynamics is poorly modeled (motion of internal organs during impact is a concern), this introduces errors that do not exist in conventional inverse dynamics.

      Bernard's list of references is very good. I would like to add one:

      Bisseling RW, Hof AL (2006) Handling of impact forces in inverse dynamics. J Biomech 39(13):2438-2444.

      They also advocate for using the same filter for kinematics and force plate.

      Ton van den Bogert

      Comment


      • #4
        Re: Should kinematics and kinetics be filtered at same frequency?

        A couple other papers on or related to this topic:

        http://www.ncbi.nlm.nih.gov/pubmed/21727008
        http://www.ncbi.nlm.nih.gov/pubmed/11747889

        Comment


        • #5
          Re: Should kinematics and kinetics be filtered at same frequency?

          Many Thanks Ton and Ross. A common concern amongst proponents of unequal filtering is essentially removing "real" forces exerted on the body. How does one rationalize through that? At first glance, they are right in thinking that real forces could be filtered out.

          Regards,
          Bernard

          Comment


          • #6
            Re: Should kinematics and kinetics be filtered at same frequency?

            Originally posted by BERNARD View Post
            Many Thanks Ton and Ross. A common concern amongst proponents of unequal filtering is essentially removing "real" forces exerted on the body. How does one rationalize through that? At first glance, they are right in thinking that real forces could be filtered out.

            Regards,
            Bernard
            A substantial chunk of real force could definitely be filtered out if GRF from something like running are smoothed at say 10-20 Hz, which raises some interesting questions on how to interpret your results if they include both GRF and joint moments and the GRF were filtered in two different ways, or what to do in the swing phase.

            A matched filter with a cutoff in the neighborhood of 10-20 Hz has always seemed like "two wrongs make a right" to me, but if it results in joint moments that are objectively more accurately than an unmatched filter without using a much more complicated model, then that's a good thing. To me the important take-home from all this is that filters should be designed thoughtfully in a non-arbitrary fashion.

            Ton touched on this a bit already but this is not only a measurement issue, it's also a modeling issue that would be important even if all the measured data had no noise at all. The Hatze paper I posted concerns this topic in a sort-of-round-about way (it's not on filtering specifically).

            Ross

            Comment


            • #7
              Re: Should kinematics and kinetics be filtered at same frequency?

              Yes, if you want to know the real force on the foot, you should use the force plate data without filtering.

              If you want to know the real force in the knee, you need to subtract m*a of the below-knee masses (which is filtered, usually) and then you will probably need to filter the force plate data to get the best possible answer.

              Or alternatively (this was in Brent Edwards paper cited by Ross, if I remember correctly) do the whole inverse dynamics without any filtering, and then filter the final result (intersegmental forces and moments) if that is needed to make it usable.

              Ton

              Comment

              Working...
              X