Now, the burning question: Where did the crack start? Over the years, in my practice, when I have a circular crack on a single fragment that someone gives me without the assembly it was in, I tend to photograph it with the location I think is the crack initiation at the 12 o’clock position. If I have both fragments, I might photograph the pair of crack surfaces with the initiations “matched up” at 3 o’clock on the left fragment and 9 o’clock on the right fragment.

One common feature that we look for to determine crack initiations is a “thumbnail” feature. This is a crescent shape that turns out to be the “natural” shape that a crack often takes as it grows from microscopic to macro scale. Imagine a tiny crack at the surface of a part. Most shafts that operate in a rotational application have the most damaging stresses running parallel to the surface. This tends to open transverse cracks. From the microscopic crack, as loads are applied to the structural component, the crack may grow. It gets both deeper and wider. Usually, it gets wider faster than it gets deeper, producing the characteristic crescent or thumbnail. This is because there are two directions in which the crack can get wider, and only one in which it can get deeper.

On the photograph in question, the dark band at the top seems to have more of the thumbnail shape than the dark band at the bottom.

Note that it is unusual for a crack to start in the interior. Surface initiations are much more common for many reasons. (See archived material on loading geometry.)

Now, what are the dark and light bands, exactly? We call them beach marks, due to the resemblance to the similar concentric curves of sand at the beach due to the action of wind and water. The edges of the bands give us a “witness” to the outline of the crack at some point in time.

My guess at this point is that the major crack initiation that grew into the “fragmenting” crack is at the top of the image.

Both of my colleagues voted for this (before I had told them about my usual choice of orientation) when I sent the image out for comments. Obviously, I probably thought the same thing when I took the photo.

However, what had me confused when I studied the photo was that cracks USUALLY merge rather than DIVERGE as they increase in length and move toward final separation. The central (near vertical, slightly jagged) ridge runs almost the entire length of the crack, but it seems more pronounced toward the bottom. The other features that had me second-guessing myself are the small flat arcs running along the circumference from 5 o’clock to 7 o’clock. But longer “meditation” leads me to think that while these arcs may indeed be crack initiations, they are most likely secondary initiations that happened only after the main crack had greatly reduced the rigidity of the shaft overall.

The fact that any way we look at it there is no obvious “final fracture” feature allowed doubt to penetrate my mind. Seeing an “obvious thumbnail initiation” and an “obvious final fracture feature” allows “epistemological confirmation” (see blog archive) of our speculation about the crack growth history. This shaft has no obvious final-fracture features such as a shear lip or final area with a helical crack that is often found on rotating shafts. So I have no choice but to give more weight to the thumbnail feature at the top. The slight upward curvature of the lowest, darkest portion of the dark lower 60% or so of the crack surface gives me a little “cover” for my increased dependence on the “thumbnail initiation at 12 o’clock” theory!        

Looking at the relative widths of the light and dark bands may also allow us, in conjunction with the (in this case made-up) records we looked up, which indeed “showed a high stress event in the machine in March of the previous year” to confirm that the final fracture in April of this year means that the dark band at the top and the light band in the center represent less crack extension than the dark (spring, warming, humid) band at the bottom. In general, if the loading remains the same over the course of the entire crack-growth timeframe, the crack-growth rate increases over time.

But why did the slight mismatch of the initial crack planes diverge as the crack grew? I am still not sure. If this was a rotating shaft, maybe there was some misalignment or intermittent impact from two closely spaced positions. I’m open to theories from people who are more familiar with the actual operations of rotating shafts.

We have now seen that looking at the crack, allowing ourselves to question the “first answer” that jumps out at us, may allow a richer stew of data from which the failure scenario may be constructed. This richer data stew is more likely to provide information that would create an effective plan to reduce the chances of subsequent failures. Note that the “data stew” comes from the failure analyst working alongside the machine operators and maintainers.