Analysis of failing parts - Mining equipment distributor

[Bruce123]

Hi everybody, This is my first time ever using a forum so I hope it works. I'm trying to work out some things in excel but having great difficulty. I work for a mining equipment distributor and I'm trying to analysis some failing parts we are having with our machines. If possible would you be able to help me on a few issues.
I have attached a file with the data and what I'm trying to accomplish.

[Wes Bucey]

Frankly, I think I would look at a root cause for the failures before implementing a statistical analysis. Are the bolts ALL performing the same function (as in a conveyor chain?) Are the bolts failing the same way (shearing? coming unscrewed from the nut or the fixture? unknown because they are popping out and lost in the mined material?)

Without an investigation into the root cause of the failure, the analysis you propose will only add numbers to a problem you already know you have; it will do little to help you reduce or eliminate the problem.

Once you know the causes (there are probably several - I can imagine the shearing or abrading on the bolts holding the teeth of the mining equipment as it digs into the workface of the substrate or bolts vibrating loose elsewhere on the equipment from the regular or random vibration), you can do a statistical analysis focusing on each of the several causes to determine a maintenance schedule to examine and replace bolts which show excessive wear, distortion prior to shearing or loose from vibration to maximize the uptime BEFORE failure puts an unexpected halt to operations.

Note there are two courses of action available:

1. establish a schedule of routine maintenance to inspect and replace bolts before failure hinders operations
2. examine the causes of failure and reduce or eliminate the number of events by using a different bolt or fastening method.

Afterthought:
It occurs to me the bolts may be sacrificial and intended to fail first before more expensive equipment fails (like a shear pin.) If so, and bolts all have same function, the question is either

1. to eliminate or reduce the excess stress which causes failure (an unlikely scenario - operators are unlikely to change mining technique) or
2. to track the events which are known to cause bolt failure and use those events as the scheduling method to inspect and replace failed bolts (like replacing air bags on cars after an accident, whether the bags deployed or not.)

[Jim Wynne]

Originally Posted by Wes Bucey

Without an investigation into the root cause of the failure, the analysis you propose will only add numbers to a problem you already know you have; it will do little to help you reduce or eliminate the problem.

That's the crux of the issue. Given the data you posted, you know that about 66% (of the failed bolts) fail within 1000 hours, and that fewer than 1% survive 1600 hours. After 1000 hours the failure rate trends downwards at about the same rate it increases prior to 1000 hours. I also notice that your "buckets" overlap; for example, 800 hours is counted in two groups.

There is some prime data missing. What sort of a population are we talking about? Your data address 108 failed bolts. If the total population is 108 (e.g., all bolts fail within 1600 hours) your strategy might be different than if the population is 10,000 or 10,000,000. The significance of failure is also important; as Wes suggests, if they're shear bolts that are expected to fail as a protective measure, it's a different problem than if the bolts are expected to perform indefinitely. It's also a different problem if, when the bolts fail, a 12-ton object falls and crushes people, as opposed to just creating an annoying inconvenience for the customer.

I also agree with Wes in advising you to (if you haven't already) ascertain why the bolts are failing in terms of what's expected of them. Premature failure of fasteners is almost always due to one of three causes: poor design of the assembly being held together by the fasteners, "bad" fasteners, or misuse of the equipment by the end user. How you approach the problem depends on the cause.

[Michael Walmsley]

Both have darn good points.

Attached is a comparison.
Data analyzed without and with 100,000 bolt population assumption
along with final comparison.

Mike