Possible FEAP bugs for triangle elements in transient problems

[Ziyu]

Dear FEAP developers,

We found a possible bug for triangle elements applied in fully transient problems. Attached are the input files and the result of Von Mises stress contour. This is simply a block under normal boundary force on its top side. No other boundary conditions applied. As you can see from the figure, some diamond blocks are littered irregularly within the block. Quadrilateral elements do not have such a mess. When triangle elements are used in quasi-static problems, the contour is also normal. I checked some source file but still get confused about this issue. Would you please give some help?

Regards,

Ziyu

[FEAP_Admin]

We checked your problem.  It works with 4,8,9-node quads and 6-node triangles.  We will have a look at the 3-node triangle to see what the problem is.

[Prof. R.L. Taylor]

The problem works fine if you request a lumped mass.  This may be accomplished by inserting the record
          MASS LUMPed

in the material definition.

There is a bug using consistent mass.  When you request a 3-node triangle feap uses one quadrature point to compute the residual.  This is fine for the stiffness terms but the mass matrix is under integrated and due to its singularity the problem is always unstable.  For other elements the quadrature use seems to avoid this singularity.

In the linear elements we do ensure a postive definite mass for the 3-node elements this needs to be inserted into the finite element too.

[Ziyu]

Hi Prof. Taylor,

I am using FEAP version 8.3. And in the subroutine fld2d1.f of this version, I think you have already fixed the singularity issue of mass matrix. You added a subroutine masst3.f, which indeed uses 3-point quadrature. Thus, how come the strange coutour still appear?

Best,

Ziyu

[Ziyu]

Hi all,

Does anyone has any other advice? I compared the displacement result computed by the linear triangle to that by quad element, finding that the displacement contour is correct. So perhaps this bug is not due to the singular matrix. It looks like FEAP just does not give a correct output stress.

Ziyu