Convergence is a universal concept in FE. If your model containes non-linearity it cannot be solved directly, so must be found be iteration. Although you cannot find an exact solution, you know you have something close when the energy you put into the model through loads roughly equals the energy output of the model through reactions.  The convergance creteria defines how close to this exact balance is acceptable.

I personally have never changed convergance creteria in ANSYS, as the program does a good job of working out what is close enough. If your model does not converge it generally means it is ill conditioned, so mesh refinement or eleminating any rigid body motion should be where you start, rather than changing the criteria.

convergence criteria is path to followed by solution to eliminate non linearities in solution

You could start by accepting the default convergence criteria which I have found to be reasonable.  Perhaps the first thing to do is to analyse a RC beam which has a known collapse load.  For example a simply supported beam under a UDL.  This might offer confidence as to whether or not ANSYS can model your problem accurately and you can then play with the convergence criteria and see how they affect your result.

You have to consider minimum increment 1E-04 for convergence criteria if you need more accurate reduce it.

I am not familiar with ANSYS. However, convergence criteria (CC) is, to my knowledge, universal concept. It is related to a iterative methods in solving PDE equation systems. Program gives estimate and based on that estimate solves equation system for current time step. If the error (difference between calculated and guessed values) is smaller than the CC, the system moves to the next time step. If the error is larger than the CC, system goes to another iteration based on improved estimation. Therefore, iterative solution will not be a good approximation to the exact solution if the CC is large - iterations are stopped too early. On the other hand, small CC requires too many iterations and wastes computation time since discretization errors exist a priori.

Optimum CC is also dependent on number of timesteps. Error induced by large CC has more chance to propagate (and reach undesirable values) through large number of timesteps.

Therefore the smaller the CC the better the solution should be. However if small changes in CC are giving you large differences in results maybe you should check your simulation model (mesh? ).

It is an iteration concept like Newton-Raphson method which reduce the non linearity to obtain optimum solution. The incremental value 1 e-4.

Convergence is a universal concept in FE. If your model containes non-linearity it cannot be solved directly, so must be found be iteration. Although you cannot find an exact solution, you know you have something close when the energy you put into the model through loads roughly equals the energy output of the model through reactions.  The convergance creteria defines how close to this exact balance is acceptable.

I personally have never changed convergance creteria in ANSYS, as the program does a good job of working out what is close enough. If your model does not converge it generally means it is ill conditioned, so mesh refinement or eleminating any rigid body motion should be where you start, rather than changing the criteria.

I agree with Phil,

when dealing with non linear simulations, each substep involves an iterative process. Usually the force is looked at as convergence criteria, in the sense that when the residual force is smaller then a force value used as criterion, then the substep is sad to be converged.

In ansys there are four convergence criteria (force, displacement, moment and rotation). When you use one of them, you may specify a value, a tolerance and a minimum reference. This last parameter is set to void a never converged situation when the value is prox to zero.

Mr. lorenzo

You said" 

Usually the force is looked at as convergence criteria, in the sense that when the residual force is smaller then a force value used as criterion, then the substep is sad to be converged."

So, how can i model the residual force in this case? I face a problem to get this part of the Load-deflection curve.

regards

convergence criteria is a rout for solution to eliminate non linearities in solution and to finalize the run.

Thanks Mr. Abbas AbdulMajeed Allawi

In problems with non unique solution it is normal to find different solutions when some numerical parameters are changed, in this case the convergence criteria. There is no correct value or parameters, if it converges to a solution all solutions are equally valid.solution more valid than others, if it converges, all solutions are correct.your

Ansys used the Newton's Raphson method for convergence by default in your problem case and there are multiple contact formulation methods. It will use Pure penalty method by default but it can be changed to MPC, Normal lagrange n other according to the nature of the problem.

Convergence is all about conservation of energy and the difference between the input energy and the work done. ANSYS is using Newton-Raphson method for "predicting" the results at each iteration and whether the result is converged or not. In my opinion it does a pretty good job on that it is not required to change the convergence parameters in most cases. Depends on the nature of the problem, convergence criteria can be considered for force or displacement in structural analysis. You can set them in "Analysis and Settings"

I agree with Phil.

recently, i performed a chip package analysis of solder ball, the constitutive model of solder material used in the model is Anand, the rest material is elastic. the model is WLCSP and is very simple, so i did not cut the model into half or quarter. I found it is had to convergence of such kind of full model. one thing has to point out that i usually prepare my model in another pre-processor: Hypermesh. i check everything in the model, there is nothing wrong in the setup. there is no contact, i.e., all match node.the boundary condition is fixed one corner, the other three corners are fixed one DOF. i also checked the materials, the units, the loads(temperature load from 200C to RT), the time steps, and even the solver advanced options. Nothing found with problems!!! So I rebuilt the model in space claim and meshed in mechanical, with the same setup. I also ask my colleague to help built another similar model using APDL. guess what, both models could convergence. So i compared the solver output files and found the force convergence criterion is a little bit different for Hypermesh model and ANSYS model, while the option for force convergence of both model is program controlled in workbench. So the i think there is maybe different convergence algorithm for external models, could anyone help explains this?