What are some of the challenges and limitations of 3D FEA and how can you overcome them?

For linear static stress analysis by FEA, solution convergence (mesh convergence) requires repeated mesh refinement and solving. This is time consuming even if it is automated. Mostly the solution convergence is generally neglected in the industry and results obtained using "industry standard meshing" are considered accurate. This problem can be solved if the mesh size can be decided before the simulation. I am writing a paper on how to decide mesh size for linear static analysis prior to the solution itself.

While this is true, I've also realised from experience that, large variations from expected result is mostly as a result of loads and boundary conditions. Most people over look this phase quite carelessly. If the mesh is fairly good, it might just be enough to setup the model until you've gotten the best BCs that represents your physicak condition well enough then you can go back and spend more time on the mesh.

Sometimes it is possible for the analyst to balance the computational effort of the model by creating breakout models (also commonly referred to as sub-models). A breakout model is when the user creates a smaller model of a specific feature they are looking to study. The smaller breakout model can have a much smaller mesh than the larger model. Since the breakout model does not contain all the features of the global model, the breakout model can have a much finer mesh without the long run-times of the full global model.

1. Mesh Quality Parameters- Aspect Ratio, Skew, Warpage, Jacobian, Taper, Max./Min. Angle, Tetra Collapse, should be maintained in an acceptable range for error free results. 2. Areas with high stress gradients should be modelled with smaller mesh size for proper study of stress flow/pattern. 3. The Order of the mesh elements should be selected based on the nature of analysis. For bending dominated problems, either Hexahedral elements or Second Order Tetrahedral Elements should be used. First Order Tetrahedral Elements will have Shear Locking effect, leading to inaccurate results.

Certain things like "walk, trot and run approach" where one can increase the complexity of the model one parameter at a time. One example could be while modeling contacts, one can start with bonded contacts and then change to more complex contact modeling in later analysis.

Sadly, almost everything we do in loads and BCs comes with approximations and assumptions, from the penalty of contacts to the rigidity of fixed BCs, to the value of loads, their direction and "conservativeness". It is very important to properly understand the physical situation we're modeling so as to be on the safe side of this assumptions. Let's not forget, FBD is a very important check after all the play.

Boundary conditions can probably be the trickiest aspect of modeling, as many other computational biomechanics experts can attest to. The most important thing to consider here is to ask oneself ' what is the application I am trying to model in real life?'

Remember that boundary conditions and material assmptions need to be validated against observed/experimental data from the system. Only then you are able to make meaningful predictions regarding the system under study.

A hugely important consideration to overcome issues in 3D FEA limitations for boundary and load conditions is to understand the Saint Venant's principle. If the region of study is far from boundary conditions and loads, you can be more flexible with constrictions and therefore obtain a good result without compromising the problem definition.

One lesson from my graduate studies (which I do not know if this is the right place to post it, but here we go...) : My geometries were so non-uniform that I had two surfaces that intersected and overlapped thus causing instabilities and crashing my model. This was not supposed to happen as it caused inverted and low quality elements. I looked back to my boundary conditions and was able to establish a contact algorithm that kept these surfaces appart thus solving my problem. I say this in order to add an alternative approach to solving instabilities in your model. In my case, the contact condition provided stability and did not affect the accuracy of my model, be very mindful of this when examining your boundary conditions.

1. A good approach while dealing with non-linear models is to add non-linearities one-by-one, this helps to identify the solution divergence source and then tweak the solver parameters. 2. Load & Displacement should be ramped over time, rather than applying them as a step input.

An effective method to overcome instabilities is to use a predefined damping factor. However, it is important to measure energy dissipation in the results to check if damping has a small effect on energy and therefore in the results obtained. If damping do not compromise results, this is an effective method to allow convergence of the system to be evaluated in silico.

The main limitation of 3D FEA is the enormous size. You should choose shell and beam elements where appropriate. The first instinct is to use 3D elements because they’re “better”, but they have no benefit in analyses where objects are long and slender. These are more appropriately represented as beams and shells.

It is critical to document the inputs and assumptions you have made when creating a model and present the full picture of this analysis for peer review. You get out of analysis what you put in, and if FEA is done simply to "check a box" then its results will not be useful nor trustworthy.

Validation is the last step in finite element analysis. After the calculations are done, this is where the user's role becomes really important. It involves processing and checking the simulation results. Even after the finite element process, we still need to validate the output. This means making sure the results are close to real-world conditions or mathematical predictions. To do this, we need to read a lot and ensure that the simulation setup has given us trustworthy and accurate data.

Verification and validation are crucial steps towards ensuring that your system replicates a real-life scenario. However, it is important to understand that models can never be fully validated, as there are assumptions made during parametrization that simplify the system of study. There are occasions where validation is impossible or very difficult because of the high costs. For these cases, a thorough verification and confirmation might provide reliable results and approximations.

Verification and validation (V&V) are essential part of any good simulation. It helps to ensure your simulation results are predicting the real-world scenario accurately. For the results validation, compare your results against some experimental testing and/or hand calculations. This would make the simulation more credible, increasing one's confidence in its results.

One semi-significant challenge in Finite Element Analysis can be how to model fastened interfaces in an assembly. How do you model the interface between the fastener (or other clamping hardware), and the members? I typically have used bar elements to represent a fastener, and other people I know have used bush elements. One challenge in both of these scenarios is that contact and friction between the two surfaces are not automatically accounted for. Therefore, all loads go through the element representing the fastener, whereas in real life much of those loads are going through the members. If contact and friction are not modelled, the analysis of the fastener should be viewed as overly conservative.