Dear all
As u know there are 6 failure patterns which the equipment or part will fail in one of those patterns Moreover , the first 3 patterns consider as age related and the others are random failure .

Can I have a real example from industrial business for each pattern

Thanks in advance

Regards,

Here is a graphic for reference to support this discussion



Terry O

Is there a way to know if the equipment I'm analyzing has a pattern B vs F before hand? I mean can we anticipate if over time we will see the increase of failures at the wear out zone (pattern B) or we can hope that the randonm failure pattern will continue stable as in pattern F.

If could anticipate the frequency of failures will increase (B), I could plan/budget the replacement sooner.

Same type of question for pattern A vs E.

Eugene, RCM2,

The six patterns are what Nowlan & Heap found applied to aircraft components at the failure mode level. They are not universal, though some or most of them seem to apply to other industries as well.

To take RCM2's questions first:
- Pattern A is the type we may expect of electrical contacts, say in motor starters.
- Pattern B is one we can expect from a sleeve bearing. We can expect 'bedding in' kind of early failures, a long period of constant hazard rate, followed by a wear out phase.
- Pattern C applies to fatigue (e.g., springs) and general corrosion (not pitting corrosion)type pf failures
- Pattern D is something I have not come across, but am told can be seen in some electrical components. Others will be able to amplify on this pattern
- Pattern E is typified by aircraft bird strikes, tire punctures by nails, pump suctions drawing in wood pieces from the sea/river etc.
- Pattern F is what applies to most complex equipment without any dominant failure modes. Even though each individual element failure mode may follow one of Patterns A-D, when we assemble these elements together, the assembly exhibits a more steady or constant failure rate after an initial bedding-in period. The battery and the engine of a car will follow Pattern A or B, but the car as a whole unit will follow Pattern F or E

quote:

Is there a way to know if the equipment I'm analyzing has a pattern B vs F before hand? I mean can we anticipate if over time we will see the increase of failures at the wear out zone (pattern B) or we can hope that the randonm failure pattern will continue stable as in pattern F.

At the failure mode level, e.g., gas turnine blade (fir-tree) root breakage, ball bearing worn, contacts welded or eroded) etc., vendors do life-trials to establish the failure probability distribution curves. If you are able to collect data and do Weibull plotting, you can also plot these distributions, but this is not a trivial task. If you do have these curves, you can determine when the wear-out phase for a single failure mode is likely to commence. Often, a
simpler way is take condition monitoring readings.

As I said earlier, most complex equipment without one or two dominant failure modes tend to follow Patterns E or F. That means there is no distinct wear-out zone for the equipment as a whole. Such items can last a long time, as long as we keep repairing or relacing the worn, eroded, corroded or fatigued parts. For these items, condition monitoring works very well in most cases.

I hope I have not gone over the top and caused confusion. If so, my apologies.

Vee: My mind tend to follow pattern F when reading your posts, confusion tends to be high at first and then get down to the green zone.

RCM,

I have a bit of a different view (actually quite different) regarding C and E in particular, but am really not keen on getting into a slanging match online here again.

Please send em an email and I will be happy to have a discussion on some of these issues in some further detail.

darylm@strategic-advantages.com

Might I suggest there are only three failure patterns if the cause is known. They are infant mortality, random and wear out. The six curves Nowlan and Heap were able to plot from the data they had would have to have been aggregations because the data did not say what was the real cause.
In looking at data for over 25 years, I have found patters that look like a two humped camels back. Now thase patterns do not fit the six that Nowlan and Heap published and does not fit a Weibull approximation either. The data in these cases is collected at componenet level and all we have is data and broken components. On futher investigation, we find three or four failure characteristics two of which are dominant mid life.
Question - can anybody name a failure mode that has a bath tub failure pattern. $100 for anyone that can.

Steve,

Good to see you on here.

I agree and also have severe doubts about whether the bathtub curve actually exists. However, I don't have the data to prove it.

Hullo Steve,

quote:

Question - can anybody name a failure mode that has a bath tub failure pattern

As you well know,we are talking about a statistical distribution, so we are not focusing on the failure of a single item, rather a family of similar(in theory identical and independant) items, or of a given item failing a number of times.

With that proviso, the example I gave earlier, namely a journal (or sleeve) bearing of the type in large diesel engines, ship propeller shafts etc., fits the bill. Some of these bearings will have early failures, partly due to bedding-in problems, and partly due to defects in material or workmanship. Those that pass this stage will run for long periods without significant failures. The failures that do occur will be due to blips in e.g., oil quality, rotor problems or other external causes. This phase will have constant hazard rates. Eventually, the babbit metal will wear out in spite of all the care. When that happens, the bearing will fail rather quickly. The time of failure of (many) bearings will bunch in a fairly tight distribution, leaduing to a high hazard rate.

You also say that there are only 3 patterns. Fatigue, fouling, general corrosion some kinds of erosion etc,. clearly follow Pattern C, which you do not acknowledge.

Eventually, what matters is to be able to decide the timing of maintenance, which is the question Eugene asked. The straight answer is that age-based intervention or PM is justified only when the failure distribution is age related, which in the N&H study accounted for just 11% of their population. For the remaining 89%, condition based maintenance offers a more optimal solution. Obviously, the actual consequences of failure will determine whether we err on the safe side with PM or are content to go with PdM. Similarly, if consequences are hidden, detective tasks are more likely to be applicable. We have already discussed RTF tasks elsewhere.

Steve >>The six curves Nowlan and Heap were able to plot from the data they had would have to have been aggregations because the data did not say what was the real cause.<<

As far as I am aware, and you may know better, United Airlines worked at the failure mode level, they knew the real physical degradation process for each mode, and they did not mix modes. Researchers of N & H's standing would have been familiar with error sources of the kind you describe. Their work was open to peer review by other experts in the Aviation Industry and the Department of Defense.

That you could not replicate these six patterns in your own studies may not prove N&H wrong. We have done similar work in the Offshore Oil & Gas Industry, and while these were not in any way comparable to the quality of work done by N&H, we did find 4 of the six patterns, with roughly similar distributions (our study involved analysis of some 30000 failures over a 17 year period on 15 Platforms. In all we studied over 900000 records and produced about 1300 sets of Weibull parameters.

Please consider whether your own example of a double humped distribution is actually caused by the blending of two distinct but overlapping failure modes.

Hi Vee... I though the provocative style of my post would get a response. Please let me respond by addressing each point in turn.

<snip>With that proviso, the example I gave earlier, namely a journal (or sleeve) bearing of the type in large diesel engines, ship propeller shafts etc., fits the bill. Some of these bearings will have early failures, partly due to bedding-in problems, and partly due to defects in material or workmanship. Those that pass this stage will run for long periods without significant failures. The failures that do occur will be due to blips in e.g., oil quality, rotor problems or other external causes. This phase will have constant hazard rates. Eventually, the babbit metal will wear out in spite of all the care. When that happens, the bearing will fail rather quickly. The time of failure of (many) bearings will bunch in a fairly tight distribution, leading to a high hazard rate. <snip>

My point exactly – three failure modes give you the bathtub

<snip> You also say that there are only 3 patterns. Fatigue, fouling, general corrosion some kinds of erosion etc,. clearly follow Pattern C, which you do not acknowledge. <snip> I do acknowledge these Vee – These are wear out or age related patterns or combinations of them

<snip>Steve >>The six curves Nowlan and Heap were able to plot from the data they had would have to have been aggregations because the data did not say what was the real cause.<<

As far as I am aware, and you may know better, United Airlines worked at the failure mode level, they knew the real physical degradation process for each mode, and they did not mix modes. Researchers of N & H's standing would have been familiar with error sources of the kind you describe. Their work was open to peer review by other experts in the Aviation Industry and the Department of Defense. <snip>

Vee (1)the practicalities of determining the contributing failure modes of a busted component are unrealistic and in many cases, (2)so is the accurate assessment of the X axis on the failure pattern. For example,
(1)a seized bearing could have had any number of knocks, lack of lubrication issues, contamination, installation deficiencies and the list goes on.
(2)Many of the causes of bearing failure are driven by starts and stops – the most aggressive time yet the bearing life would have been determined by engine hours if the bearing was installed in an engine. Here we would introduce a skew towards random failure as bearing life would be highly influenced by the aircraft operating context – short haul or long haul.

Now the investigator who logged the failed bearing probably did not have the time to diagnose which was the dominant failure and even if he/she did, what they have is a bearing that is likely to be severely damaged and in most cases to the extent such that an accurate determination of the failure cause would be difficult to ascertain with certainty. The likely scenario in United Airlines was that the person who replaced the bearing threw it away without much of a look…. and could have logged the fault at gearbox or shaft level.
I have read the original Nowlan and Heap report but not studied the data. I am basing my proposition on my experiences. When I worked in aviation, we raised defect reports on a small percentage of failures and took some photographs occasionally but that was about the extent of the investigation unless there were unacceptable trends with certain components or some serious fleet wide reliability problems. The reality of the Nowlan and Heap work is likely that they had lots of data and were able to produce graphs, but realistically I can see no way that they would be able to tease out the failure modes with any precision.

<snip>That you could not replicate these six patterns in your own studies may not prove N&H wrong. We have done similar work in the Offshore Oil & Gas Industry, and while these were not in any way comparable to the quality of work done by N&H, we did find 4 of the six patterns, with roughly similar distributions (our study involved analysis of some 30000 failures over a 17 year period on 15 Platforms. In all we studied over 900000 records and produced about 1300 sets of Weibull parameters. <snip>

Vee, I am in no way saying Nowlan and Heap are wrong They found from the data they had that it could be categorised into six unique patterns – I am saying that people often make unsupportable extrapolations from this work. Sure I get failure data that looks like a bathtub curve but what this tells me is there are at least two failure modes working and my data is unable to separate them,

<snip> Please consider whether your own example of a double humped distribution is actually caused by the blending of two distinct but overlapping failure modes. <snip>

This was the truth when we did further analysis. One of the cases whas a mechanical anti skid unit on an HS748 aircraft. Two dominant problems - spring fatigue and seal failure that let grease onto the clutch mechanism. All we had was data at anti skid unit level. The overhaul facility showed us the problems and we were able to make some assessments after this but this was a significant amount of work - not the type of analysis that would have been done on each component Nowlan and Heap would have had data for.

This is a very interesting topic - I have had these discussions many times before. This is a good forum topic that often gets people thinking.

Thanks for the interest Vee
Regards
Steve
www.reliabilityassurance.com

quote:

Originally posted by Ozgipsy:
I agree and also have severe doubts about whether the bathtub curve actually exists. However, I don't have the data to prove it.

Daryl, the bathtub curve is live and well. We recently did some work for a refinery where the failure pattern on the main air blower exhaust fans was a bathtub. We find this pattern frequently.

Steve,


But not at a failure mode level.

When you collate these things up and use them at the equipment level of course they will turn out like this.

Correct Daryl.. The bathtub curve does not exist at failure mode level. The bath tub in the fan belts were installation problems and belt stretch and the wear out was wear but interestingly the wear out had a wide variance caused by corroded pulleys accelerating wear. There were three failure modes in reality.
Please see the power point presentation - notethat users should click the slide to see the information unfold

MainAirBlowerPresentationfanlife.ppt (68 Kb, 63 downloads)

Steve,

You say,

quote:

My point exactly – three failure modes give you the bathtub

What 3 failure modes are these? I see only one, bearing worn or siezed. What we can see is there are 3 time slices where some of the sample bearings fail due to different causes (degradation mechanisms). The cause of failure is not the same as the failure mode, or am I missing something here? I dont think I am on the same page as you are, so help me out.

Hi Vee,
Sorry to take so long to get back to you. Weekend etc. I have been looking for a definition of failure mode as this is the core of the problem. When I was taught and the way we teach is that a failure mode includes the mechanism. Your definition is different. SAE JA1011 defines Failure Mode as follows:
A single event, which causes a functional failure.
I can see your reasoning now and perhaps you can see mine.
On the subject of defninitions, I think the mechansims needs to be included in the failure mode because if there are three mechanisms operating each one needs to be evaluated. There could be three defensive strategies deployed and in RCM, the way to do this would be to have three failure modes listed.
In your analysis, do you list your failure modes and your mechanisms independently?

Steve,

You say
>> that a failure mode includes the mechanism.<<

Earlier you said,

quote:

For example,
(1)a seized bearing could have had any number of knocks, lack of lubrication issues, contamination, installation deficiencies and the list goes on.
(2)Many of the causes of bearing failure are driven by starts and stops – the most aggressive time yet the bearing life would have been determined by engine hours if the bearing was installed in an engine.

I agree entirely with the inclusion of mechanisms in the failure mode definition, so that we define exactly how an item fails. I do not dispute that an impeller worn is not the same as impeller blocked.

Your earlier post highlights the practical difficulties of extracting such differences from the records, e.g., in a CMMS or operating log or work order. Taking your siezed bearing example, I would argue that it is very difficult in real life to know how many were due to contamination, installation etc. So they do get lumped together, as 'bearing siezed'. I would go one step further and add 'bearing worn' to it, as in most cases the work order and history would simply record it as 'bearing damaged and replaced'. Further, bearings do sometimes sieze if they are worn. I am all for recording data accurately; but in all my working life, I have rarely, if ever, found such precision in the records.

To come back to 'time slices', consider,
1. Journal bearing has early life failure due to poor lub oil (i.e., cause is recorded)
2. Journal bearing has random failure due to poor lub oil (i.e., same cause)
3. Journal bearing has wear-out or old-age failure due to poor lub oil (i.e., same cause)

Stating the cause does not identify the time-slice. It only helps identify what to do about it. We are no better in knowing the differences' in as you call it, the 3 failure modes. In my view there is only one failure mode, whether you state the cause or not. If we are going to base our decisions on actuarial analysis, not vendor's life-tests in simulated conditions, we are stuck with making these approximations. I am not an academic, and have no time for esoteric theory we cannot apply in practice. At the same time, I do not scoff at theory, it is very useful in improving performance. But that should be our sole objective, to improve performance. Just look around and see how many RCM practitioners do any kind of reliability analysis at all in deciding the 'when' of maintenance tasks and you will appreciate my point.

Vee,
I agree... I think we got stuck on a definition of failure mode. Data collection is retrospective and it is difficult to assign a failure mechanism but RCM is futuristic and we contemplate what may occur in real life.
My definition of failure mode always includes the mechanism whether it is practical to allocate the mechanism or not.
The practical side of me says that a bathtub must be a combination of at least two mechanisms hence in my definition two failure modes as there would have been two mechanisms.
If you are personally happy with the failure mode definition of seized then you can get a bathtub curve with one failure mode.
I think that the SAE standard definitions supports the view that the mechanism is included as it talks about a single event which would say there are two or more failure modes operating in the bathtub.
As we know, maintenance has many words which are defined and used differently by different people. Here it seems we have another case of this.

I would like to hear from "RCM2" and Eugene. Have Vee and I confused you. I think Vee gave you some examples. If you have any questions still, please punch em in.

Hi Steve. I'm cool. I like to start this type of discussions or pitch in with a question or comment to spin the participation of knowledgeable members like Vee; and then sit back to enjoy reading and learn.

As I said to Vee; my mind has an "F" curve but now I have reached the green zone.

Just my two cents, I have seen the bathtub and also curve A.
Steve one of your considerations was installation problems. I had the "luxury" to work in a brand new plant with "fresh" operators and mechanics on new (call it unknown) equipment. Guess what, infant mortality high, now we are several years further and wiser and things are "easier' right now. But on some equipment it is starting to rise. Probably if we phase some of the troublesome equipment out, I suspect that we will fall more in the category A, wiser from past mistakes.

The plant would rather follow pattern F

Just sitting on the same bench, next to Eugene