One of the benefits of having seasoned engineers doing design work is that they have an innate ability to know when a structure “feels strong.” They have designed and built similar structures in the past and know where problem areas are, when something needs extra support or if a selected part “looks too thin.” They also know when something looks questionable; and that’s when they request an FEA be done.
FEA stands for Finite Element Analysis (yes, saying FEA Analysis is like saying ATM Machine). FEA works by using a computer to break a complex part or structure into very small pieces. The computer takes imputed conditions and calculates how the model will respond. For mechanical engineers, FEA is used for structural testing: looking for overstressed areas, finding excessive bending and discovering areas to save weight. It has its origins in the aerospace industry, where stresses, flexing and weight savings are a daily concern. FEA can also perform heat transfer, fluid dynamics, earthquake testing, vibration studies and many other applied fields.
As a mechanical engineer for STS Technical Services, I use FEA for design validation; determining if a proposed design is strong enough to take standard and non-standard use. Our client will request a design safety factor, standard and emergency loads and lifetime requirements, which I will use as inputs to perform the FEA. With the results, problem areas can be found and solutions to strengthening the design can be found.
FEA is strongest when it is backed-up with physical testing. A typical development cycle without FEA might go through 10 prototype cycles: design, test and redesign. Using FEA, development can be reduced to one or two cycles, drastically cutting costs.
A common mistake is to take FEA results as a guarantee that a design is bulletproof. An analysis is only as good as its inputs. The more sophisticated the model, and the more time taken to accurately constrain and load a model, the better the results. FEA is at its best when two models can be directly compared – a 20% improvement between similarly prepared models is much more reliable than the stress reading from a single study.