I am working on a simulation of a sintering process to predict contractions and distortions associated with densification. I am currently using a custom UMAT subroutine implementing a viscoplastic law, and the analysis is performed using the *Visco procedure.
Model Description:
Setup: A component is placed on a rigid base (defined as a *RIGID BODY) under gravity loading (*DLOAD, label GRAV).
Step Type: Non-linear analysis with *VISCO and NLGEOM.
Contact Definition: I am using Mortar Contact (TYPE=MORTAR on *CONTACT PAIR), as it has provided the best results so far.
Surface Interaction: High friction coefficient (~0.4). My ideal target is PRESSURE-OVERCLOSURE=HARD, though I have experimented with others.
The Problem: I am facing recurrent convergence issues during the contact phase.
Geometry dependence: Convergence is extremely difficult in cases with point-like supports, but even geometries with flat supports are generating issues.
Contact Stiffness: I have tried using PRESSURE-OVERCLOSURE=LINEAR with very low contact stiffness. Given that the sample is only under gravity, I have used values significantly lower than the recommended 5x to 50x Young’s modulus, which helps slightly but results in extremely slow calculations.
Damping: I am aware that *CONTACT DAMPING is only available for implicit dynamic calculations and not for static or visco procedures like mine.
Controls: I have attempted to relax the convergence criteria using the *CONTROLS card (adjusting FIELD, CONTACT, and LINE SEARCH parameters), but this has not resolved the instability.
I am looking for suggestions to improve convergence and stabilize contact in this type of analysis. Are there specific Mortar settings or alternative stabilization strategies for *VISCO steps that I might be missing?
Attached is an example .inp file (ViscoAnalisis INP) representing the model features. Please note that I cannot share the proprietary UMAT code, but the rest of the setup is there for visualization.
Did you try with a built-in creep law too ? Did you try different values of the *VISCO, CETOL parameter ? You can find some tips for the built-in creep model here: Creep calculation and results
I have bad experience when it comes to convergence with Mortar contact, but sometimes it may work better than penalty contact.
Your model may have initial rigid body motions and is loaded with force which is bad for convergence. It would be best to remove advanced features one by one to see exactly what causes convergence issues.
but as i understand, Mortar contact is for hard type purpose to achieved better result and convergences
did it mean Mortar contact does not work for many case of your models, but another contact type works? can problem sketch or benchmarking with another solver provided since my experiences shown Mortar contact work better in many case e.g clip snap fit were Abaqus failed in hard contact type.
In those first increments where I was facing more troubles, the viscoplasticity was not active yet (it’s activated when a certain temperature is reached), so i don’t expect the cetol parameter makes an effect on that initial convergence.
Thanks for the suggestion. I agree that rigid body motions and force-controlled loading can affect convergence, and I have been simplifying the model to identify the cause.
I achieved a significant improvement in convergence by changing the rigid base elements from C3D20 to C3D8R, while keeping the same rigid body constraint and contact setup. Even though the base is rigid, the element formulation still affects how the contact problem is solved numerically. The simpler C3D8R elements led to a more stable and better-conditioned contact behavior, which noticeably improved convergence.
I recall a few cases where user’s model had convergence or penetration issues and changing from standard penalty to Mortart contact only made it worse and it failed to converge pretty much immediately. If I find a specific example like this, I will share it.
Indeed, second-order elements are known for potential issues in contact:
all quadratic elements cause problems in contact calculations, because the nodal forces in the vertex nodes equivalent to constant pressure on an element side (section ) are zero or have the opposite sign of those in the midside nodes
Full integration is also bad for materials with high Poisson’s ratio or models with significant yielding.
indeed, Mortar contact has frequently in problem at early implementation, temporary disable or remove also in previous but later being improved then. Thanks, sharing and reporting some of your problems is important for attention to user and developer probably.
regarding element type, i like faster in modeling and meshing so there’s tetra with quadratic requirement. I did not find any significant problems with surface to surface type of. contact, CalculiX run successfully and convergence faster even the case of large multipart contact.