In the 1970’s and early 1980’s mills experienced “overall” structural failures. Over time, analysis and design methods have improved, and the vast majority of failures we see today are “local” failures on the mill structure. These are often caused by stress concentrations of the geometry, such as at flange endings, or other local discontinuities. Sometimes they are caused by occurrences of rare local load combinations, which may not have been considered in design. For example, if a mill is stopped by a brake, acting on a flange integral with a segmented head, on rare occasions the brake may grip the flange just at the end of the segment. In that case, the resulting brake forces will not distribute, circumferentially, to both sides of the brake calipers, but only to one side, due to the flange split line. This produces a different load pattern than the expected typical.

Recently there have been some mill failures, on the shell mounted mills, that deviate from the norm. Since the shell cans (without horizontal split flanges) are axisymmetric, it is expected that, on a circular cross-section, every point will see the same stress range. Thus, if we had a circumferential weld, the expectation of a failure crack starting, would nominally be the same for every point on the weld circumference. Indeed that WAS the occurrence on the early “overall” failures. Yet several more recent mills, of the shell mounted design, have experienced circumferential weld failures where the cracking was very severe (through the thickness) in a local arc area, but the rest of the weld circumference showed no signs of cracking [1]. Also, this cracking occurred in only one of a group of similar mills, but different sites showed the same failure patterns.

To explain this, one needs to consider what possible non-symmetry can occur (ruling out a rash of “drop charges” at the various installations):

Weld flaw distribution need not be axisymmetric, since QA would not differentiate flaws which just passed a specification from flaws which passed by an order of magnitude. But this would not be unique to shell mounted configurations.
Machining one side of a shell plate, like a riding ring, could create thickness non-symmetry, if the fabricated shape were not circular to start. This would be unique to shell mounted designs.
In welding without stress relief, the residual stresses do not have to be symmetric. This would be true for field or shop welding.
Some non-symmetric distortion is specific to field welding. This is usually measured, and corrected in the field, but with less ease and accuracy than in a machine shop.

Finally one needs to consider shoe bearing design. These bearings are designed for self adjustment. That being so, they can tolerate more rotating structure deformation, yet still operate satisfactorily. But in doing so, they introduce non-symmetric load variations, if a deformation, like a local flat spot, passes over the bearings. Also one needs to consider the effect of friction in these shoe bearing components, as they adjust.

The above is pure speculation, but it does show that there are some non-symmetric influences, and combinations, in shell mounted construction, that may not exist in trunion designs. Since all these are variable for each individual mill, they could explain both the non-symmetric nature of the failures, and why only a few, out of a batch, failed. The answer as to whether any of these speculated effects are really significant, awaits a suitably designed mill instrumentation program.

[1] Svalbonas, V., and Schultz, K., “The Need for Strain Gaging in Mill Design”, CONMINUTEK, April, 2013, Iquique.

Addendum (January 31, 2016): The above discussion tacitly assumes the failure initiates in the weld. The weld, being the weakest point in the design, and at a location of highest stress, seems to be the likeliest failure initiation area. However, a recent investigation into an unrelated, fabricated gear rim failure, initiated another line of conjecture.

The riding ring undergoes some significant residual stress, during the fabrication sequences of NDE, rolling, welding, and more NDE. Suppose that these residual stresses combine with undiscovered flaw areas in the riding ring plate. This would create an arbitrary weak point(s) in the riding ring.

The figure above (left and right) shows an identical failure pattern on two different, shell supported mills, at two different mine sites. As noted before, it is natural to assume that the problem started in the weaker weld zone, where it cracked completely through the structure. But then, if the crack starts in the central weld zone, how does one explain that the crack growth does not continue along the weaker circumferential weld, but instead travels into the stronger riding ring parent plate, at both ends? An alternative explanation is possible. The cracks could START at the flaw points in the rolled riding ring plate (perhaps enhanced by points 2 and 3 in the previous discussion), and then meet up, traveling along the weld. While the weld is still the weakest area, and thus cracks through, the problem initiation points are at riding ring plate flaws. This explains also the non symmetry of the failure, since initiation is not dependent on weld stresses.

How is the riding ring plate examined? This is usually done by straight beam UT, which would miss planar flaws oriented through the plate thickness. During rolling, the induced stresses would act on any of these planar flaws oriented perpendicular to the rolling direction. This would create a weak zone along the direction of the observed crack ends, in the figure. While this is by no means a complete explanation, it does point to a possible failure mode, in shell supported mills, which may not currently be considered. The low probability (but not zero) of such flaws, also explains why only a few mills fail.