Sometimes it seems “pretty obvious” why a failure happened. If you look at a burst pipe, for example, maybe the edges of the split area seem pretty thin, maybe much thinner than the adjacent material. So maybe it seems obvious that the wall got thin and could no longer hold the pressure. But how do you KNOW whether the wall thinned down BEFORE the burst or as a RESULT of the burst? Maybe it burst because the pressure increased unexpectedly.

One time I was working on a broken semi trailer axle. One of our lab technicians helped me to unload the fragments from the pallet in the shipping area. As it happens, he used to be a truck driver. To him, “the answer” was OBVIOUS! “HE JAGGED IT!” the technician pronounced. In other words, the driver jack-knifed the truck, which caused the axle to break. OBVIOUS! To him! However, the axle was never on a truck! This was a durability test specimen that was loaded in fatigue until it broke. There was actually nothing wrong with it. It lasted three times the minimum requirement of the manufacturer. It was broken. This analysis was being done to see the detailed metallurgical characteristics of an axle that performed WELL!

Recently, I worked on a petroleum-product storage tank that leaked. One of the people who inspected the tank noticed that there was weld spatter on the outside of the tank. Well, that looks bad. The hole that allowed the contents to leak was very round. Maybe it was due to a defect caused by some weld spatter that had subsequently been rusted away along with the surrounding base material. Good. Is the analysis over? Well, maybe. Sometimes all that is allowed in a failure analysis is a visual inspection. However, in this case, cutting the leak area to allow inspection of the inside revealed multiple other partial through-wall pits. Clearly the more likely answer was that the corrosion was a result of internal corrosion rather than a weld-spatter defect.

Sometimes it is not at all obvious why something cracks. Several of my clients make plastic products that are reinforced with metal inserts. When the hot plastic is molded around the cooler metal, the plastic shrinks more on its way down to room temperature than the metal. This is what is generally used to hold the metal in place in the plastic. Sometimes knurls or grooves or other geometric features are used to enhance the gripping power of the plastic so that the metal part stays in place. But, the differential shrinkage ends up causing the plastic to be stretched. Whenever materials are stretched, there might be potential for a crack to form. In this case, it is usually obvious that there is a tensile stress, and the question usually is something like “Why did THIS ONE crack?” The answer is often complicated. If the plastic was a little warmer than usual it might shrink more on cooling, leaving a greater tensile stress. Or maybe the geometric “retention features” (knurls, etc.) were sharper than usual! Or maybe there was some contaminant, like the metalworking lubricant used to manufacture the insert, that caused “environmental stress cracking” in the plastic. Or maybe the plastic got contaminated with some other grade that was not as crack resistant. I have found cases over the years for all of these reasons.

One of my clients decided to get fancy and do some modern engineering work. They actually got a finite element analysis (FEA) model made that predicted the relative shrinkage of the plastic and the metal. To support the model, they had a lab measure the thermal expansion coefficients for the plastic and the metal. They sent the lab data to the people who were making the model. Well, when I looked at the parts and looked at the model, the high stress area shown by the model had no relationship to the actual crack area. That might be a hint that there is something wrong with the model, the data or that somehow the assumptions that were made about the parts were not representative of the actual parts. I looked over the model, but to be forthcoming, I am not enough of an expert to really be able to tell if the modelers did something wrong - unless it is really gross. Nothing jumped out at me as being suspicious. Then I looked at the data. Somehow the lab had come up with thermal expansion coefficients for the metal that were higher than those of the plastic! This seemed VERY STRANGE to me. I never heard the end of this story, but on researching the specific materials in question, SOME OF THE DATA HAD TO BE WRONG.

Too many people today try to use advanced engineering techniques, such as model building, without taking the time to think about whether the results make sense. We have powerful tools today to help us do our jobs better. An FEA model can be very helpful in understanding WHY something cracked. If the stresses in ANY single location approach the STRENGTH of the material, a crack can be initiated. Whether the stresses are due to differential cooling, applied loads, inertial effects created by high loading rates or even a corrosion process, models can be made to shed light on a given situation. But, as a wise man said years ago, when my son asked him, “What does the Bible tell us?” “It doesn’t tell us anything. We have to study it to know what its message is.” The same is true of a finite-element model. It is just a model. If we checked our brains at the door when we came into work, it is not going to help us understand our situation.

More recently, someone brought me an FEA model of some parts that had been unexpectedly cracking in a durability test. So I looked at the pretty color map, and asked what it showed. Well, it was showing the stresses on the surface of the part. They were “von Mises stresses.” Great. Von Mises stresses show the amount of localized shear stresses. Since the cracks in question were fatigue cracks, this was good. Fatigue cracks generally initiate in shear, but once the crack is initiated, NORMAL stresses (for a review of shear and normal stresses go back to Obvious and Not So Obvious Ductile and Brittle Fractures: Part 1 by Debbie Aliya originally posted on this blog July 7, 2008) are what usually drive the crack to grow the fastest.

So if you have a crack, you need to understand the NORMAL stress situation. If you have a fatigue crack, you also need the shear stresses, and Von Mises is one way to get that. So the FEA model in this case had not been examined very thoroughly once it was made. FURTHERMORE, the model was a LINEAR model. That assumes that there is no area of the component that reaches the yield strength. Well, in order to have a fatigue crack, at least one grain must be stressed above the yield strength. So a linear FEA will only give a VERY rough starting idea of what is going on. In many cases, the stresses are highest on the surfaces, so starting out with a stress map showing only the surface may be a reasonable approximation.

To get the most out of the usually considerable effort required to make a good FEA model in the first place, make sure whoever is making the model understands what you want to use it for, and make sure that the results seem to at least QUALITATIVELY show what is really happening. If they don’t, something needs further attention!

CONCLUSION: A properly designed and executed finite-element analysis can shed light on why a failure happened and can be especially useful as a supplement to a failure analysis when there are a lot of competing factors of unknown relative importance that MIGHT have facilitated the failure.