All Simulation Packages use FEM 1

[GandalfTheGrey]

Fellow FE Analysts,
I came across this article by SolidWorks which has put me in dilemma.

SolidWorks Featured Paper

It says the accuracy of all Simulation packages are same and depends on material, boundary conditions, etc. to which I agree.
However, I know that in ANSYS, we have two solvers viz. Direct Sparse Matrix Solver and Iterative PCG Solver (Conjugate Gradient Method combined with a Pre-conditioner), midas NFX uses the SA (smoothed aggregation) AMG (algebraic multi-grid) method.
Theoretically, a problem can also be solved by Galerkin or any other variational method.
Wouldn’t the accuracy depend on the solver technique used for SolidWorks?

Please provide your valuable opinion.

 

[GregLocock]

"It says the accuracy of all Simulation packages are same" Not what it says. I hope your models are more accurate than your reading skills.

"many of these packages utilize the same underlying mathematical
approach: the finite element method (FEM). In other words, the accuracy of simulation results is
not dependent on the underlying mathematical method, which generally is the same, and has a
lot more to do with whether a problem is set up correctly. "

I'm not arguing with that. I think it is a bit misleading but not wrong. An important difference between solvers is the speed with which they get to within a certain accuracy, and the skill with which the programmers have optimised that.



Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[FEA way]

Indeed structural analysis software generally uses FEM (this is not true for CFD where FVM is more common and more accurate). However, as you say, they have different solvers. Not only different types of solvers but also coded in another way. For simple benchmarks it might be hard to notice differences in results but you will always see some discrepancy (that’s why these benchmarks are used after all). For more complex real-life problems the differences become even more visible. This is caused by the fact that even though basics of FEM are the same, more complex features are handled by each program in distinct way. This is especially true when it comes to contact formulations and some other features used in nonlinear analyses (not to mention special-purpose techniquees such as XFEM).

 

[GandalfTheGrey]

Thanks FEA way.
That makes sense.

 

[Mustaine3]

When I use in iterative or the direct solver in Abaqus, then the criteria for the residuum are the same. The difference is just how the tangent stiffness for each iteration is created. This also counts for other methods.

The main question is, how strict a tool checks the residuum (so the out-of-balance forces) and when a solution is accepted or more iterations are necessary.

I've already compared results from CATIA V5, Solidworks and Abaqus. When the same things are defined in a linear static analysis, then the results are the same.
Small difference in results are more likely to come from different settings in result visualization (averaging, etc.).

 

[karachun]

If different methods used in proper way, they should give almost same results.
For example, Frequency response problem can be solved using direct method or modal superposition. Modal superposition method claimed as less accurate but if user request enough modes then difference in result will be neglectable.
Many linear static solvers use direct sparse solver as default and iterative solver as option for large models that cannot be fitted into memory. Results of iterative solver will be close to sparse solver if user set appropriate residual value.
My opinion is that all solution methods have reasonable mathematical accuracy if used in prober way.
Modeling errors like bad quality mesh, wrong considerations like ignoring large deflections in structure where you should account them, and real-life factors like manufacturing defects are greater sources of inaccuracy than FEA math.

 

[SWComposites]

It says the accuracy of all Simulation packages are same and depends on material, boundary conditions, etc. - that statement is pure uninformed dangerous rubbish.

 

[GregLocock]

Just thinking about it - every professional package I have ever used has a bug list. Unless they all have the same bugs then they must produce different results.

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[ESPcomposites]

Also, some are more forgiving than others. For example, a shell element can be relatively complex and exhibit hourglassing or shear locking. The allowable skew angle or aspect ratio is also a function of the formulation (and not just the order). Another example is something like an Eigen solver, where different algorithms (and checks for them) can be more or less robust than others (some may have trouble with negative Eigenvalues or miss some). There may also be internal settings to the program (such as tolerance levels, etc.) that may not be easily adjusted (or can not be adjusted). If those are not set properly, odd things can happen with little or no warning. One should understand that these program are extremely complex and it is much more than directly solving an equation that has single (and exact) answer. Many of these issues have been addressed over the decades (or improved upon) by the high end commercial packages, but it is not a given for every solver (especially for packages that are lower cost and less history).

Brian

http://www.structuralfea.com

 

[ThomasH]

I would say thet the "article" is more like a broschure describing how good Solidworks is. I think you can find a similar broschure for Abaqus at Dassault Systems site.

My experience is that different solvers can be good (fast) for different problems. Some solve a dense matrix fast, others a sparse. There is a lot of variables that the three page broschure ignores and simply quotes a satisfied user claiming that Solid Works is much faster that Ansys for them.

I don't see anything new in this, it's advertising.

Thomas

 

[IceBreakerSours]

More often than not, a model's ultimate value has more to do with the analyst than the tool.

Accuracy without a well-defined context is meaningless; it is useful term for marketeers but, sadly, too many engineers fall for it. No numerical tool is "accurate" for all scenarios; it simply cannot be. I have tested multiple commercially well-known structural solvers on the same set of nonlinear problems with default settings and found them giving more or less the same answer; it was the cost/performance/.. that separated them.

*********************************************************
Are you new to this forum? If so, please read these FAQs:

http://www.eng-tips.com/faqs.cfm?fid=376

http://www.eng-tips.com/faqs.cfm?fid=1083

 

[izax1]

Intersting discuasionQ! But how do we define "accurate"? I bet that no matter how accurate you set up your FEM model, it will never copy the reality. And reality is also relative depending om how well you manufacture the part and how accurate the system is assembled. After all, FEM is only a numerical approximation and the theory can be 100% correct, but still wrong compared to reality.