Tuesday, October 05, 2010

How Much Theory You Need To Know

    My colleagues and I are disturbed when we hear companies say their software brings finite-element analysis down from the ivory towers inhabited by Ph.D. gurus. Most professional stress analysts do not have a Ph.D.

    In our division, we have more than 90 engineering analysts, and only about 20% have a Ph.D. Some of the best analysts I know have only a B.S. degree, but what makes them good is a track record extending over 20 years or more. When I interview job candidates, I focus on how well they can learn, not how many letters they have after their name. It is true that many professional analysts have an M.S., but it is not because they need to know tensor calculus. Instead, the fact they have pursued an advanced degree demonstrates they can learn. A doctorate is additional evidence they can do original thinking.

    I think thoughtful analysts are afraid of pushbutton analysis, but not becuase it might cost them their jobs. There are still many crucial problems that pushbutton approaches cannot solve. So there will always be work for experts. What analysts actually fear are errors that can be built automatically into an analysis at the push of a button, sometimes in an uncontrollable and even undetectable fashion.

    One company claims its automesher will build models with an element form sometimes superior to conventional tetrahedrons. But deep inside the model, these elements look as though they are near the event horizon of a black hole. This affects the answer. Many engineers, and most salesmen, lack the two or three years of differential equations, numerical analysis, and vector calculus to understand why.

    One widely used solver is pitched as a product that makes analysis accessible to just about anyone who can type. We use this package because it is an excellent tool for large 3D linear elastic stress analysis. But no matter how good your executive assistants are, would you want them designing airplane parts?

    The only way to design critical parts safely is to use your head - and somebody else's as well. If you can't tell approximately what the answer should be before you statr a problem, ask someone who can. Be sure you do this before taking the word of a computer. Many a neophyte has been saved by a graybeard saying: "That looks like it will break. Add material here."

    Analysts should be as familiar with Roark and Peterson's books as they are with their favorite FEA text or user manual. We have on occasion seen "answers" that violate Newton's Laws of Physics. These errors apparently were tied to users who didn't understand the physics of the problem. We have also seen symmetrical problems with unsymmetrical answers. The software vendors involved couldn't explain the errors. In both scenarios, we were saved from the "answer" by people with experience who instinctively knew something was wrong.

    Many engineers have the background and experience to use pushbutton analysis safely. But many in the target market do not. The notion that an academic degree in engineering instantly qualifies you to be an analyst is equivalent to saying that a new driver's license qualifies you to compete in the Indy 500.

    Finally, bear in mind that a good stress analyst is like a good trial lawyer. Neither asks a question for which he does not already know the answer.

by Mr. Douglas Hall

No comments: