This really depends on the nature of the problem. A failure (fracture, wear, deformation) of a component in service requires detective work to determine if the component was abused or exposed to any unexpected conditions, and to determine if it was manufactured correctly per specifications. First, observe everything while altering as little as possible. Look at fracture surfaces for signs of fatigue, tempering oxide (which would indicate a flaw that existed during or as a result of heat treatment), etc. Chemistry should definitely be checked to make sure the part was made to specifications, but an accurate chemistry measurement is typically destructive. You can screen by XRF or other methods to non-destructively determine if the material at least has the correct alloying elements to possibly be what it is specified to be, but getting the carbon level of steel within the few hundredths of a percent of variability that is allowed in most types of steel, for example, requires destructive testing of a sampling of the material. There are texts that outline the steps of failure analysis in detail; these are just examples of the first few things.
If it is more of a design issue where an existing material is thought to be inadequate, you would still want to verify that the existing components were being made to print. Maybe the design was adequate, but a vendor has substituted materials without getting approval, or some part of the process is out of control and no one is aware. Once that is verified, I would look at what types of failures are occurring. If certain types of failures occur consistently, there is less detective work. Even if the parts are being "abused", if every user is "abusing" them, the manufacturer might need to redefine what they consider to be abusive. Either users need more guidance and warnings, or the product has to be better. Again, the failure mode is important. Brittle failures have different solutions than ductile or wear related failures.
I wouldn't really consider a completely new design to be a "problem" but you might start with a particular material based on past experience and find out it isn't adequate, which would than lead to my previous paragraph. In addition, you have to realize that lab testing might not perfectly reproduce real world conditions. You could be producing failures in the lab that would never be reproduced in actual use, and actual use might produce failures that you can't reproduce in a lab. The problem can be the tests rather than the material or design.
------------------------------
Kenneth Kirby
Senior Project Engineer
Snap-on, Inc.
Kenosha WI
(262) 748-3836
------------------------------