Dangers of Modern FEA
Hang around any engineering office for more than a few days and invariably you will see Finite Element Analysis (FEA) results. The widespread growth of this technology has put unprecedented analysis power in the hands of engineers to solve problems, however it has not come without difficulties.
Over the last decade FEA tools have become increasing user-friendly. The graphical user interfaces have dramatically improved and the use of these powerful tools has become significantly more intuitive. While this has made the technology far more approachable for rank and file engineers, it has not come without a cost. Some say that the increased approachability of the tools has led to a decrease in the quality of engineering being performed using it. Ask engineering managers about their impression of FEA results and you will be greeted with suspicion. I have many times heard “…we tried FEA before but the results weren’t that good. They don’t tend to match up with what really happens.”
Back in college when I first began using FEA, it was a complicated mess. You had to individually pick element formulations, map contact nodes, choose solver types, and manually perform validations of the solution steps. It was a time-intensive chore to produce an FE model, and due to the computing power available, you had to be judicious with your choices to maximize the validity of our models. Because of this, we spent a great deal of time researching element types and solution paths to determine the strengths and weaknesses of each.
Modern FEA tools are far simpler to use. Years ago when VGO purchased our first seat of Ansys Mechanical with Workbench I was excited to try out the new interface. I built a simple cantilever beam model, applied boundary conditions, meshed it, and solved the model all within a few minutes. With just a couple of clicks and virtually no training on the new interface I had a pretty picture on the screen. I felt happy at how quick and easy it was, and was impressed by how far the technology had come. After a few short moments the joy I felt began to fade and worry encroached as I started to ask myself simple questions.
What material is this using? I realized that the model solved without my ever inputting a material definition. A traditional FEA code would produce an error if I forgot to input material properties (and another error if my Poisson ratio, shear modulus, and Young’s modulus were inconsistent). I began searching through the help files to find where the material should be specified and eventually came across the fact that the code is “helpful” and chooses a default material for me. In this case it choose “structural steel”, a generic steel with properties similar to what one may expect for a low carbon steel. While on the one hand it was nice that the FE code very much wants my solution to run without errors (and it took steps to ensure that), on the other hand I was somewhat bothered by the idea that it made a very critical decision without consulting me. This began to get me concerned about what other decisions the code made for me in the name of making it easy to use, and “error” free.
What type of mesh did it use? The mesh, or the discretization of the solution space, it of critical importance in FEA solutions. The types of elements, aspect ratios, densities, number across critical areas etc., all substantially effect both the validity of the results and the specifies of the question that is being asked by the model. In my first model by clicking the “mesh” button, the program applied a tetrahedron mesh on my cantilever beam. Not necessarily a bad thing, but this mesh type has some issues with bending. In this case, the automatic mesher used only one element across the bending thickness, which is well known to poorly simulate bending. Now I want to point out that I did have complete control over the mesh – and the mesher is very powerful and gives me the freedom to make the mesh in any way I see fit. The concern is that it was very easy for me to use the automatic controls which in this case resulted in a mesh unsuited for my application. If I hadn’t known better, I would have assumed that it was OK based on the ease with which the process progressed.
What element type is it using? Within each meshed element shape, there are many element formulations that can be used. These element formulation include the mathematical shape functions and behaviors that the element exhibits. In this case, a general purpose structural solid element formulation was chosen (Solid 187 if memory holds). While this was appropriate for my model, it was difficult to find the information on which formulation had been used, and as I subsequently learned it was a bit difficult to change. In traditional FEA tools you had to manually select the element type, which forced you to consider the pros and cons of your choice.
What other decisions is it making for me? This is just a short list of a couple of the real obvious decisions that Ansys made for me that day. The main point is that if I didn’t have a good working understanding of how FEA works, I could easily assume that my answer was accurate simple because it appeared straight-forward to navigate a simple looking user interface and produce a result that looks like other models I have seen.
Automation is becoming more and more a part of daily life, and that’s a good thing, freeing us up to focus on larger questions; but we should be aware that not all of the decisions that automation is making are good ones for our particular need. Just like any engineering tool, we need to learn about how it works, and the limitations that exist, such that we don’t draw the wrong conclusions from a “pretty picture”.
I encourage engineers to learn about this powerful tool, as it opens up a whole world of solution paths not previously available to us. Here is a brief list of suggestions I would make to help with the process of learning how to use FEA effectively.
1. Learn how FEA works – In engineering we are used to closed form solutions to the equations that we use on a daily basis. FEA is not a closed form solution. Learn the fundamentals of how it arrives at a solution – it may surprise you.
2. Learn the basic limitations and assumptions – There are many types of FEA models; get a basic understanding of the main ones: linear elastic, geometric non-linearity, material non-linearity, axisymmetric, plane strain, plain stress, linear buckling, and modal analysis. Just learning about the differences will give you a powerful head start in knowing what you can (and can’t) do with FEA.
3. Always cross check with hand-calcs – No matter what type of engineering analysis you are doing you want to cross check with known solutions. This is especially true in FEA where it may not always be clear how to properly input loads and constraints. Realize that there may not be a one-to-one with your FE model, but you want to draw commonalties in behavior and magnitude to confirm you are solving the problem you think you are.
4. Do your convergence studies – If you aren’t doing convergence studies, you can’t be sure that your solution is mathematically valid. Valid doesn’t mean that it’s correct, but at least it has arrived at a solution. Without knowing convergence you are guaranteed that you don’t have the answer to the question you asked (regardless of if you asked the right question).
5. Make sure you are asking the right question – Each FEA model is designed to answer a specific question. Even if your model is valid, it may have nothing to do with the problem you are trying to solve. Early on, ask lots of questions, look at a problem from many viewpoints. As you get experienced, you will gain a sixth-sense as to what questions are probably important.
6. There is no such thing as a rigid boundary condition – The hardest thing about FEA is getting boundary conditions (BC’s) that simulate your problem. The easiest BC to apply is the fixed or rigid condition – unfortunately it never exists in nature. Play with changing boundary conditions and you will quickly see just how significant this is.
7. There is no “Perfect Model” – I will get some cabbage thrown at me for this one. Some analysts believe that if you perfectly model everything then you can get all your results from one model. This is not the case because perfect does not exist in the real world. Geometries are not perfect and loads are not perfect (or repeatable). What you want to do is determine what aspects are important, and that takes many what-if scenarios. Once you have completed your modeling tasks, you will understand how the system acts and what causes it to behave the way it does. Now you’re in a strong position to effect positive change.
8. Make lots of models – People always ask me how long it takes to make an FEA model. I usually tell them that the last model I run takes maybe 5 minutes, but it took the previous 20 models for me to figure out the right way to constrain the model and what aspects it was sensitive to. Until you make many models with perturbations you only know how that one model behaves. Try slightly changing dimensions, angles, load points, etc. You will find that some models are very sensitive to minor changes. Think of FEA as a process that once completed educates you about the system – it’s that education that is valuable – not an individual model.
9. Sensitivity studies are your friend – The most important thing in FEA models is the behavior of the system – how it is acting. You will find that if you get the behavior right, everything else falls into place. For every BC you apply (loads, constraints, etc.) try doing it several different ways (point loads, pressure loads, contact applied loads, etc.); if your desired output (stress, deflection, etc.) significantly changes than you need to work to make sure your decisions simulate the actual conditions; if you find the results is insensitive to your changes, then you can be less precise with that parameter.
10. Be careful with CAD based FEA tools – The trend of bringing analysis software to the engineer’s desk has resulted in a situation where all major CAD programs include “analysis” capabilities in their software. These are essentially highly limited versions of FEA software. They are great for what they are, but they potentially exacerbate the problems discussed above. Many of them automate almost all of the actions, to the point of removing the engineer from many decisions. Not necessarily a bad thing, but you run the risk of getting a “pretty picture” that has nothing to do with your problem – and you won’t know it. Cross checking becomes MUCH more important, and make sure you cross check with known problems very similar to yours.
11. Be careful with your results – Be sure you output the results that are germane to your question. For example, FEA gives us the power to output stress in a number of ways. You can ask for Max Prin., Min Prin., von-Mises, shear stresses, tensors (at an orientation) principle directions and others. It is very common that von-Mises is shown by an analyst, even when it isn’t appropriate to the problem being solved. In future posts I will talk more about von-Mises “stress” and which results you actually want to use for predicting various types of failures. Hint: von-Mises has nothing to do with cracking failures or fatigue – it’s not even a stress. Look it up.
Feel free to call or email with questions,
Dave Van Dyke