MultiBody.Joints.Constraints.UniversalJoint uses == on Real

[modelica-trac-importer]

Reported by jmattsson on 25 Feb 2013 08:57 UTC
Modelica.Mechanical.MultiBody.Joints.Constraints.UniversalJoint contains the equations:

      // Constraint equations concerning rotations
      if n_a[2]==1 and n_b[3]==1 then
        // Rotation order 1-2-3, 1 locked
        R_rel.T[3,2]=0;
        frame_a.t[2]=0;
        frame_b.t[3]=0;
        elseif n_a[3]==1 and n_b[2]==1 then
         // Rotation order 1-3-2, 1 locked
        R_rel.T[2,3]=0;
        frame_a.t[3]=0;
        frame_b.t[2]=0;
        elseif n_a[1]==1 and n_b[2]==1 then
         // Rotation order 3-1-2, 3 locked
        R_rel.T[2,1]=0;
        frame_a.t[1]=0;
        frame_b.t[2]=0;
        elseif n_a[2]==1 and n_b[1]==1 then
         // Rotation order 3-2-1, 3 locked
        R_rel.T[1,2]=0;
        frame_a.t[2]=0;
        frame_b.t[1]=0;
        elseif n_a[1]==1 and n_b[3]==1 then
         // Rotation order 2-1-3, 2 locked
        R_rel.T[3,1]=0;
        frame_a.t[1]=0;
        frame_b.t[3]=0;
        elseif n_a[3]==1 and n_b[1]==1 then
         // Rotation order 2-3-1, 2 locked
        R_rel.T[1,3]=0;
        frame_a.t[3]=0;
        frame_b.t[1]=0;
      else
        // No rotation
        {1,1,1}=R_rel.T * {1,1,1};
      end if;

The n_a and n_b vectors are of type Real, and the specification clearly disallows == and <> for Reals.

The rationale given in the specification for this limitation mostly concerns variables, so it might be reasonable to lift the restriction for parameters.

---

Migrated-From: https://trac.modelica.org/Modelica/ticket/1002

[modelica-trac-importer]

Comment by sjoelund.se on 25 Feb 2013 10:44 UTC
I looked into why OpenModelica doesn't care about this. And yes, we only make the check for variables. For structural parameter there is no issue at run-time. And even for non-structural parameters... During initialization, there are no state events, right? ;)

If MSL wants to follow the specification, the code should be re-written or we need to schedule this for spec 3.2rev2. I'd propose the spec changes.

[modelica-trac-importer]

Comment by hansolsson on 25 Feb 2013 12:48 UTC
The model(s) needs to be rewritten.

The underlying reason for disallowing comparison between real variables is that one needs to consider that they are not exactly computed; this can occur during simulation but also due to the imprecise calculations for parameters.

In this case it is even more problematic(*) - there are two real vectors as input and only 6 combinations of values are supported for them (and then an else treating all other cases the same), i.e.

 if n_a[2]==1 and n_b[3]==1 then
    // Rotation order 1-2-3, 1 locked
    R_rel.T[3,2]=0;
    frame_a.t[2]=0;
    frame_b.t[3]=0;
   ...
  else
    // No rotation
    {1,1,1}=R_rel.T * {1,1,1};
  end if;

If you only support a limited set of values you use an integer or enumeration instead of testing for specific values of reals; that clearly documents it for users and avoids the problems.

The same goes for Constraints.RevoluteJoint
*: I haven't svn updated today in case it was recently changed

[modelica-trac-importer]

Comment by hansolsson on 27 Feb 2013 08:30 UTC
Looking more it seems the correct solution is to remove all of the code above and replace it with something like (note that n_a and n_b have Evaluate=true so for the cases above it should generate the same result):

frame_a.t*n_a=0;
frame_b.t*n_b=0;
n_b*R_rel.T*n_a=0;
assert(abs(n_a*n_b)>Modelica.Constants.eps, "The axis must be different");

Assuming I did not misunderstand something.

[modelica-trac-importer]

Comment by hansolsson on 27 Feb 2013 08:33 UTC
Except that the assert should be on something like norm(cross(n_a,n_b)) instead of on abs(n_a*n_b) - or possibly assert that abs(n_a*n_b)<...

[modelica-trac-importer]

Comment by otter on 3 Mar 2013 13:19 UTC
Fixed by Andreas Heckmann (the author) in 392e03f

[modelica-trac-importer]

Comment by dietmarw on 3 Mar 2013 14:09 UTC
392e03f introduced this:

  //Modelica.Constants.eps 

why? Since its commented out it should never have been added.