The usage of "Gearing" in FDK

Main Forum
1

Hi everybody.

I have been building a knee model using FDK for the past 6 months or so.
In this time frame, a frequently occurring error was:
"Failed to resolve force-dependent kinematics. Too small steps for GSS search causes looping of Newton method"

I have consulted the following webpage.

In here, it recommends for me to set the ForceTol to 1 N, and adjust the MaxNewtonStep accordingly.
In addition, since the "final force error" that occurred in the simulation is either N or Nm, it was recommended to use gearing (referred to as AnyKinMeasureLinComb RotGearing on the webpage) for the rotational degrees.

I couldn't understand how gearing is supposed to work. Is it supposed to alter the MaxNewtonStep so that the actual max for the rotational degrees is described as Gearing x MaxNewtonStep?
If so, I don't understand how this would affect the ForceTol of that particular degree of freedom.

It would be great if someone could explain how gearing works.
Thank you in advance.

Kazuki

2

Hi Kazuki,

It is a good question the gearing concept can be a bit abstract.

Here is an example:

  • You have a linear DOF we can call it "lin"

  • If you control the position of this DOF with FDK the ForceTol and MaxNewtonStep meaning is straight forward, it is N and m respectively

  • Now assume that you make a AnyKinLinComb measure we call it "lin_geared" using the "lin" measure, and that you make it so that the output of the linear combination measure is a gearing of lin so that "lin_geared=lin*gearing" .

  • Now instead of using the "lin" dof for the FDK we use the "lin_geared" for the FDK, this will indirectly change the meaning of both ForceTol and MaxNewtonStep.

  • This is because that the FDK solver will now work on the "lin_geared" and when the solver change this measure the "lin" measure will be changed more or less depending on the gearing, this will indirectly change the meaning of ForceTol and MaxNewtonStep.

In my experience it can be important to change the gearing so that the sensitivities for the different DOF are more or less of the same scale of magnitude, but be aware that the ForceTol may be either increased or decreased depending on the gearing.

It can also be an option to think in different formulations of the DOF so that you express the rotations using linear constraints.

Please also note that the surface quality can impact the convergence, so try to make a small test like described here https://github.com/AnyBody/support/wiki/All-about-Force-Dependent-Kinematics#anyforcesurfacecontact

Best regards
Søren

3

This topic was automatically closed 125 days after the last reply. New replies are no longer allowed.