J. Known Issues

The following is a list of known issues in Sierra/SM as of the most recent code release.

Section 4.3: Deactivation of element blocks (see Section 6.1.7.9) does not currently work in conjunction with the full tangent preconditioner in Adagio. To use this capability, one of the nodal preconditioners must be used.

Section 6.2.1: For problems with large rotations, the hourglass energies are known to spike exponentially with increases in total rotation. This behavior is observed for both midpoint and strongly objective strain incrementation and for both incremental and total hourglass formulations. The large rotation issue is a known limitation in solid mechanics for all non-hyperelastic material models.

Section 6.2.13: For problems with rotations, the SPH algorithm generates rotational spring like forces on the boundary because of known deficiencies in the algorithm in distinguishing between rigid body rotation and deformation. The SPH algorithm does not conserve rotational or angular momentum.

Section 6.2.14: Superelements are not compatible with several modeling capabilities. They cannot be used with element death. They cannot be used with node-based, power method, or Lanczos critical time step estimation methods. They are also not compatible with some preconditioners (such as FETI) for implicit solutions.

Section 7.4.2.2: If a prescribed displacement with the CYLINDRICAL AXIS option is applied to nodes that fall on the axis, it will have no effect. Separate boundary conditions should be applied to those nodes to fix them in the plane normal to the axis.

Section 7.4.3.2: If a prescribed velocity with the CYLINDRICAL AXIS option is applied to nodes that fall on the axis, it will have no effect. Separate boundary conditions should be applied to those nodes to fix them in the plane normal to the axis.

Section 8.9.6: Attempting to use GLOBAL SEARCH INCREMENT with a value greater than 1, especially in a problem that contains shells and/or restart, will, in most cases, cause code failure. A GLOBAL SEARCH INCREMENT value greater than 1 will, under the best circumstances, give only a marginal improvement in speed.

Section 9.7: User defined variables (see Section 10.2.4) are not currently supported with heartbeat output.

Section 8: Contact of one rigid body against another is currently inexact. This includes contact of actual meshed rigid bodies, analytic surfaces, or incompressible lofted objects (particles, contact through the thickness of shells, or contact through the radius of beams). As these contact systems have no elastic response, the contact solver may need many iterations to solve these systems or may fail due to an unsolvable system. This could lead to energy conservation issues in the result of contact between these objects.

Section 4.3: Although multiple linear-solver packages are available for use as preconditioners for solving implicit solid mechanics problems, we strongly recommend that the FETI equation solver or plain CG be used in production analyses. The FETI equation solver is actively maintained and tested by the code development team. The FETI EQUATION SOLVER command block is documented in Section 4.4. Other solvers may not be thoroughly tested and may have compatibility problems with some code capabilities such as MPCs and contact.

Many of the more advanced constraint types are not fully compatible with one another. Thus if a single node is involved in multiple types of constraints the resultant behavior may be unstable, particularly if the constraints are not fully orthogonal. Currently supported types of constraints, involve kinematic (such as prescribed velocity), rigid body, periodic, MPC, contact, and spot weld. Kinematic constraints are compatible with contact. MPC constraints are somewhat compatible with contact. However, most other combinations have limited compatibility. Thus for example, a node with both a contact constraint and a periodic constraint may lead to model instability.

Epic Material Model Issues

Epic (MMM) materials will have problems when used with several of the following capabilities.

  • Converting MMM elements to SPH particles

  • Given a set of MMM SPH particles adding (via particle conversion) or subtracting (via element death) from that set. Changing the contents of the particle set will cause a redistribution of volume among the integration points, not just addition or subtraction of volume at the changed particle.

  • Nodal based tetrahedra, and non-local continuum elements have the same non-local issue as SPH particles.

  • Mapping of a solution from one mesh to a different mesh will cause a similar problem.

The problem with the above is that many of the epic material models explicitly store the integration point volume as a material state value; the change of this volume is used to compute an incremental strain rate for equation of state calculations. Any method that yields changes in material point volume unassociated with material deformation will cause these models to have severe accuracy and stability problems. If, for instance, on conversion to particles, a new particle returns a volume of 90% of the old element, the material interprets this as a 10% volumetric compression over a single nanoseconds-long time step; the material can view this as a monstrously large shock. This can cause the material point to introduce an immense amount of artificial energy to the system and drive the model immediately unstable.

In order to avoid these problems when converting MMM materials to a particles, convert them to a standard material type (such as elastic plastic). Additionally do not use the MMM materials when using any of the non-local or mesh dynamic methods listed above.

SPH Particle Method Time Step Stability Issues

Many particle problems that are unstable at the default time step can be made stable if they are run at a manually reduced time step. There are a number of related issues: SPH time step calculation is not that robust to begin with, particularly in high volumetric compression. If over a single time step, two particles can move more that about one half particle radius length, which can cause an instability. Effectively, this will cause the SPH material to invert. This can be an issue with high velocity impact problems. The relative particle velocities are not taken into account in the critical time step calculation, so the code has no knowledge when this may be occurring.

If this problem occurs, it may be fixed by dropping the global time step scaling factor (say from 1.0 to 0.75 to 0.5) and examining if stability can be obtained. Generally if stability cannot be obtained at a factor of 0.5, there is something other than time step driving the instability.

Particle Creation Issues

If two SPH particles happen to be created extremely close to one another (for example, less than 10% of a particle radius), the particles will fly apart violently, adding significant artificial energy to the system. The closer the particles are, the large the energy will be. If two particles are created exactly on top of one another, the error is infinite. This excessively close particle creation can happen with particle conversion or embedded particles. If the source elements for the particles are heavily deformed and in close proximity, new particles can be created close by to existing particles. Additionally contact and/or particle repulsion may be imperfect in high velocity impact cases, allowing bodies to temporarily occupy the same space. If a newly converted particle happens to be placed on top of an existing particle, a major influx of energy may ensue, causing instability.

Possible input deck fixes include:

  • reducing the time step,

  • reducing the potential for contact or particle repulsion errors,

  • adding a “failsafe” death criterion to detect and remove particles that have gone unstable (for example, when these types of errors occur, the problem elements often have astronomically high velocities. Establishing a death criterion of velocity that exceeds, for instance, 10 times the initial impact velocity may detect and immediately remove the problem elements, preventing the stability error from propagating to the rest of the analysis.), or

  • creating more robustly enforced contact (the use of more contact momentum balance iterations may help prevent materials—and thus newly created particles—from occupying the same space).

Bad Element Shapes Yielding Invalid Contact Geometry

Elements will continue to return some material response as long as they maintain positive volume. However, long before the point of absolute element inversion, it is possible for the contact geometry of elements to become corrupted. For example concave and locally inverted elements may have exterior faces that point the wrong way. An incorrectly pointed face can cause major contact errors to occur either adding artificial energy to the system or preventing solver convergence. These types of extreme deformation can be common in high velocity impacts. Use of a nodal_jacobian_ratio <= 0.0 death criterion can help avoid this issue. For eight-noded hexahedral elements, the contact geometry for an element is guaranteed to be valid at least until this shape metric reduces to zero.