Solver Settings for Contact Analysis
This section provides useful tips for tuning solver settings for contact analysis in general, and tips specific to different methods.
General Settings
The following solver settings can help to successfully perform contact simulations in general:
•
In a contact analysis, you almost invariably use an incremental approach. It is possible to solve a problem without friction in a single stationary load step, but such an approach will often fail to converge. In a stationary analysis, you should instead use the parametric continuation solver, and gradually increase the load or displacement. Enable it by selecting Auxiliary sweep under Study Extensions in the settings for the Stationary solver.
•
Use a direct solver, rather than an iterative solver, as linear equation solver if the problem size allows it. Direct solvers are less sensitive and can provide better convergence.
•
As a default, the double dogleg nonlinear solver is selected when a stationary study is generated and Contact nodes are present in the model. For the majority of contact problems this solver has more stable convergence properties than the Newton solver, which is the default solver for most other problems. Using otherwise similar settings, the double dogleg solver tends to be somewhat slower than the Newton solver on problems where both solvers converge. It is, however, often possible to take larger parameter steps when using the double dogleg solver. For some problems, the Newton solver can still be the better choice, so if you experience problems using the default settings, try to switch solvers.
•
For contact problems, it is often necessary to let the parametric solver use a defensive strategy when going from one parameter step to the next. This can be controlled by setting the value of Predictor in the Parametric. By default, the parametric solver will do so by setting the predictor to Constant when contact is present. However, it can sometimes be more efficient to use a more aggressive strategy by setting it to Linear.
•
Always solve contact problems that involve friction or decohesion incrementally, using a parametric or time-dependent solver. The evolution of the friction forces is history dependent. For contact problems without friction, an incremental strategy is not necessary but often a good choice.
•
If the model includes friction, try solving the problem without friction first if possible. When the study runs absent friction, enable friction again.
•
The convergence of many contact problems can be improved by modifying the parameter or time step algorithm. For a stationary study, you can tune the step size in the Parametric node, and for a time-dependent study, you can modify the time stepping in the Time-Dependent Solver.
•
For models that include decohesion, see the suggestions under The Decohesion Node.
The Penalty Method
Since the penalty method only adds a weak contribution to the physics, there are no special solver requirements, apart from the suggestions that apply to contact analysis in general. If the simulation shows poor convergence, or even diverges, it is for the penalty method often related to the choice of the penalty factor. A too high penalty factor may cause the ill-conditioning, which can manifest itself either in poor convergence of the Newton solver or large errors in the linear solver. If this is encountered, consider using a smaller penalty factor. However, a too small penalty factor will deteriorate the accuracy of the simulation, and contact might even be lost if the overclosure becomes too large.
The Segregated Augmented Lagrangian Method
The segregated solution method for the augmented Lagrangian formulation introduces extra contact degrees of freedom for the contact pressure and the friction forces. It also relies on a special solver sequence to obtain the correct solution. Some general considerations when using this method are:
•
The convergence check relies on the scaling of the degrees of freedom, but since contact pressures and friction forces often are zero over parts of the simulation, you should not rely on automatic scaling. When the solver sequence is first created, both contact pressure and friction forces are given a manual scaling which is relevant for typical metal-to-metal contact. You should in most cases change this to values appropriate for your application. The variable scaling is accessed under Dependent Variables in the solver sequence. Set the scale for each variable to a value that is representative for the expected result. Too large values may give a too early convergence, while too small values may lead to an excessive number of iterations.
•
The default solver sequence generates one lumped step in the segregated solver for each Contact node. This split of variables into different lumped steps does not influence the solution as such; you can equally well group the contact variables in a single lumped step. Each lumped step will however generate an individual curve in the convergence plot, making it easier to pinpoint the source of possible convergence problems. You can also increase the granularity even more by changing Solver log to Detailed in the Advanced node in the solver sequence. This will give a separate convergence curve for each dependent variable.
The default solver generates a solver sequence that is stable and gives and accurate solution for the majority of contact problems. However, if convergence is difficult other settings can be tested:
•
The Segregated solver is generated with a termination technique set to Tolerance. This setting ensures the convergence of the contact degrees of freedom, that is, that their value only changes within the specified tolerance during the latest segregated iteration. If, however, convergence of the contact DOFs is of less importance, you can accept the solution after n segregated iterations instead. To do this, set the termination technique to Iterations or tolerance and specify the maximum number of segregated iterations. The solver will then continue to the next step if the tolerance criteria is fulfilled, or if the maximum number of iterations is reached. Note that the solution of the segregated groups is still converged in each segregated iteration.
•
For efficiency purposes, the nonlinear solver in the Segregated Step that includes the displacement field is by default set to accept the solution after seven iterations, regardless of convergence or not. If you notice in the solver log that the solution is far from convergence after these seven iterations it can be necessary to change this setting. Updating the contact DOFs with a nonconverged solution can cause the overall problem to diverge. By changing the termination technique to Tolerance, the segregated solver will instead do a cutback if such a situation is encountered.
The Coupled Augmented Lagrangian Method
The coupled solution method for the augmented Lagrangian formulation introduces special contact degrees of freedom for the contact pressure and the friction forces. Otherwise it places no special restriction on the solver sequence. Some general considerations when using this method are:
•
The convergence check relies on the scaling of the degrees of freedom, but since contact pressures and friction forces often are zero over parts of the simulation, you should not rely on automatic scaling. When the solver sequence is first created, both contact pressure and friction forces are given a manual scaling typical for metal-to-metal contact. You should in most cases change this to values appropriate for your application. The variable scaling is accessed under Dependent Variables in the solver sequence.
•
Since the solution to the augmented Lagrangian can be nonsmooth, the default double dogleg nonlinear solver in stationary studies is sometimes too conservative. The convergence can in such cases often be improved by using a Newton solver, for example, the Constant (Newton) with a full Jacobian update.
•
Although the penalty factor does not affect the accuracy of the solution, it can have significant influence on the convergence properties of the model.
The Nitsche Method
Since the Nitsche method only adds a weak contribution to the physics, there are no special solver requirements, apart from the suggestions that apply to contact analysis in general. If the simulation shows poor convergence, or even diverges, it is often related to the choice of the penalty factor. A too low value may cause the solution to be unstable, while as for the penalty method, a too high value may cause ill-conditioning.
Dynamic Contact
The dynamic contact methods in general inherit the properties and suggested solver settings from the corresponding standard method. With regards to solver settings, the main difference for the dynamic methods is how to set up the time-dependent solver, where it is recommended to use manual time stepping, see Dynamic Contact Analysis.