|When Good Engineers Deliver Bad FEA
By: PAUL KUROWSKI
President, Acom Consulting,Toronto, Ontario, Canada (article plugged from http://machinedesign.com/bde/cadcam/bdecad3/bdecad3_5.html)
There is no doubt that finite-element analysis is getting a bigger role in development projects. One reason is that it helps slash expensive prototype testing. The technology is also seen as another way to improve product integrity.
Despite FEA's reputation for accurately pinpointing weak spots in designs, a few faulty assumptions and organizational flaws may render analysis work unusable. For instance, some companies treat FEA as an extension to CAD packages. In fact, it requires specialized training all its own.
To better spot lapses in how companies implement FEA, we've tallied a list of hazards and faulty assumptions. Each of the problems cited can have serious consequences.
What could possibly go wrong?
Most vendors claim their FEA programs are easy to use, implying that almost anybody can become an instant FEA expert. If this is true, why not ask CAD departments to handle FEA, since they already have access to it as a part of their CAD systems? The flaw in this thinking is the assumption that FEA is an extension to CAD.
FEA software is getting easier to use, but structural analysis by FE methods definitely is not an extension to CAD. Similarities between CAD and FEA are superficial, and limited at best to geometry. Proficiency in CAD does not assure expertise in FEA, and someone skilled in running an FEA program is not necessarily a good analyst. CAD operators running FE programs often strive for the most accurate representation of geometry because geometry is the prime focus of drafting. Meshing, element type, loads, supports, error estimation, and result analysis get less attention.
Another rash assumption is that model accuracy equates with precise geometric representation, and that one can assess model quality based on visual appearance. This is reinforced by vendors pushing their CAD-FEA interfaces for "better geometric accuracy and easy meshing". Also, vendor-provided training often focuses on geometry and other easy CAD-like functions.
Unfortunately, FEA is not that easy. Geometric accuracy and impressive color plots do not equate with a good model.
|Assigning FEA to engineers lacking product experience....
Assigning FEA to engineers lacking product experience produces another hazard. To produce meaningful FEA results, the analyst must know the principles underlying the finite-element method. The analyst also needs practical experience, a feel for design, and sound engineering judgment. He or she must understand the product as well as its intended work environment. This knowledge is critical for tasks such as deciding which features must be accurately modeled, deleting or simplifying others, determining how to apply loads and restrain the model, analyzing errors, and conveying results back to the designer.
Assigning FEA operations to recent graduates produces other problems. Newcomers may not be comfortable interacting with others and withdraw into an isolated world of computer simulations. This situation does not contribute to the learning process that young engineers need nor does it serve the best interests of the engineering department.
The question persists: Should design engineers perform FEA themselves or should we have specialized FEA personnel? The best approach is to have design engineers run the analysis because they know the product. Design engineers with proper FEA training could perform analysis interactively while designing, as long as they use software intended for this mode of use. Pro/Mechanica (formerly from Rasna Corp. which is now part of Parametric Technology Corp.) is one example of software that integrates well with design work. More complicated analyses, such as nonlinear problems, can still be farmed out but in tight collaboration with the designer.
Too narrow a job description produces another FEA pitfall. Even when there are dedicated personnel for design analysis, the FEA-specialist job description should be abandoned. When FEA is someone's life, that person uses FEA on everything. The technology gets the nod even when hand calculations or physical testing would be faster, less expensive, and more accurate. FEA should be interactively used during design, prototyping, testing, and product follow-up. The FEA analyst should have a hand in those activities as well.
Looking for the instant specialist is also a bad policy. Some job ads explicitly ask for experience with particular software, again indicating a superficial, CAD-like approach to FEA. Understanding the FE method is more important than specific software commands which are easily learned.
Not enough finite-element methods training time produces another hazard. Quite often, the only training comes from software vendors. Such instruction may be superficial and concentrate only on how to run software rather than on understanding FEA. A person eager to use newly acquired software skills and lacking a good grasp of FEA is probably the most dangerous user.
How to be a smart FEA shopper
Demos and training often portray vendor-supplied software in the most favorable light while simultaneously concealing shortcomings. Here are a few guidelines for shopping FEA software:
Poorly defined FEA objectives lead to wasted effort. When the analyst receives a project requirement saying only "perform FEA on the control arm," then objectives need work. Goals should cover why the analysis is needed, the expectations, and how results will be used. The perform-FEA syndrome often stems from bureaucratic misunderstanding rather than engineering need for the results.
Lack of project monitoring leads to time-and-cost overruns. Standard checks should take place during an FEA project to monitor progress and provide guidance for analysts. This helps the manager follow the project and quickly spot problems. Failing that, mistakes may never be uncovered.
No lessons-learned database means mistakes are often repeated. Each project should be well documented so third parties can recreate results long after the analyst is gone. A sample of completed results should be confirmed through testing. Where there are discrepancies, an appendix to the project report should address the problem. Users need to verify results with experiments until they get confidence in the method.
An FEA report should be self-explanatory and contain enough information to duplicate analysis results. A good report together with backup, provides sufficient detail for rerunning the analysis without any additional instructions.
No real commitment to FEA is an attribute of managers with a short attention span who become disappointed quickly. Building confidence in the method, and accumulating and maintaining in-house expertise, takes years of considerable effort and commitment. Nobody should expect instant savings.
Once introduced, FEA is considered an omnipotent method to assure quality designs. But a few unsuccessful application attempts can make people give up not realizing that the discipline failed simply because of lack of a quality assurance system.
|Here's a test for your manager of analysis
The four models in this study represent the same bracket. It is rigidly supported at the back and loaded with uniform pressure applied to the top of the hollow cantilever. Each model produces different results. Which one is correct? One cannot tell, based on an examination of meshes. The answer is hidden in the formulation of each model.
Model 1 produces Von Mises stress of 18,000 psi.
It uses a first-order solid tetrahedral element which, by design, can only model constant stress within its volume. Knowing that, there are two big problems. First, only one element is placed across the thickness of the plate in bending. This model is not capable of representing bending stress which changes from compressive to tensile across the plate thickness. Consequently, bending stresses are badly represented by constant stress.
The second problem is that the elements are highly distorted. Each type of element works well only if it is within specified shape limits. If element distortion is beyond these limits, then numerical procedures used to calculate displacements and stresses return false results.
The two problems represent the most common abuse of FEA. Either problem renders this model useless and potentially dangerous, depending on the function of the bracket.
The mesh on model 2 is similar to that on model 1 but uses second-order solid tetrahedral elements. These can model linear stress distributions within their volumes. However, the mesh is too coarse to model stress distribution correctly or detect stress concentrations. Some elements are still highly distorted. The maximum Von Mises stress is greatly underestimated again making this model either useless or dangerous.
|Model 3 produces maximum Von Mises stress of 49,000 psi.
This model uses second-order solid tetrahedral elements and has enough elements to model stress distributions properly. Elements are also correctly shaped. However, there is no way to know if the mesh is fine enough to produce reliable stress information. Model 3 is a starting point for good analysis but now needs several mesh refinements to examine stress convergence and estimate solution error. Stress will increase with each mesh refinement. Thus, the process of mesh refining and solving the refined model must continue until the increase in stress between two consecutive iterations becomes sufficiently small. Only then can results be accepted as final.
Unlike the preceding models, this one uses adaptive-order elements or p-elements from the Pro/Mechanica code. With p elements, the solution proceeds through several automatic iterations while element order upgrades as needed to satisfy user defined accuracy. In this case, the accuracy gage is less than 5% error on local strain energy, local displacements, and global RMS stress.
As a result of the adaptive elements, the analysis solves successfully despite having only one p-element across the plate in bending. Also, p-elements can accept much larger distortion and are easier to use with automatic mesh generation. What resembles a distorted h-element is still a good p-element. The model needs no further work if 5% accuracy is sufficient.
|ISO 9001 sets FEA categories
Formal guidelines for an analysis program are found in ISO9001 Document R0013. It also ranks FEA use into three areas.
Category 1: Vital
Failure will put human life in danger or cause a public disaster. FEA is an integral tool proving product integrity. Products that exemplify this category include aircraft, pipelines, bridges, and dams.
Category 2: Important
Failure may put human life in danger or cause serious damage. The product failure may be vital in the sense of Category 1, but FEA is not used exclusively to demonstrate product integrity. Product examples include cars and home appliances.
Category 3: Advisory
This includes all situations not covered by categories 1 and 2. FEA contributes significantly to product integrity. Failure would cause only financial loss.