Sponsored by the Association for Maintenance Professionals Join or Manage Your Profile Posting Boards Machinery Condition Monitoring and Predictive Maintenance Posts About Technologies and Techniques for Condition Monitoring Weibull - old thread - new stories?
Join or Manage Your Profile Posting Boards Machinery Condition Monitoring and Predictive Maintenance Posts About Technologies and Techniques for Condition Monitoring Weibull - old thread - new stories?|
Go
|
New
|
Find
|
Notify
|
Tools
|
Reply
|
|
We've had discussions regarding Weibull Analyses previously; but???? Either there wasn't much ado with it regarding actually getting the job done or.
Question: is there case histories where this process has actually improved plant maintenance or machinery condition? Cordially, Sam |
|||
|
Doesn't the New Weibull book contain case histories? Also can check some examples in www.barringer1.com to see whether it contain things that you require.
Perhaps many know it's important and useful but not trained to apply it. |
||||
|
Sam,
I led a team which has done a large volume (35000 failures, 900000 records)of Weibull analysis several years ago. The major problem we encountered then was one we are all very familiar with, namely data quality. At that time our CMMS had history in a text field, so using some clever software, we could locate a lot of the failures. With ERMs and coded entries, errors at source are very difficult, if not impossible, to resolve. In my view, it does not help doing sophisticated analysis with source data of indifferent quality. No amount of fancy footwork can hide the fact that the dancers are flat footed! As to whether the Weibull analysis was useful, it did a few good things. First it proved that the failure patterns in the Airline Industry were largely repeated in our Industry too, with slightly different distributions. We could use the results in RCM and other Risk analysis. The search software helped us locate failure history records rapidly (nothing to do with Weibull itself, but we would never have acquired the search software if we were not doing Weibull). It showed us how poor our data recording was, and that led to some initiatives. The exercise led us to analyze Fire, Gas, Smoke and Line-of-Sight detector performance., as well as that of ESD valves, Pressure Relief Valves etc. This extension of the study came in very handy when discussing longer test intervals with the Regulator. My suggestion to all readers is to focus on data quality first. a number of things have to be in place to enable high data quality, some are listed below. - Drop down lists in coding fields should contain at most 5 items. The option 'Other' should be avoided or at least minimized. - While some class room training is necwessary, it is not sufficient. Mentoring and one-to-one coaching is essential. Most errors arise when people don't know what to do when there is a mandatory field which will stop them from closing the screen. Put in any rubbish and the problem (for them) is solved! - Minimize manadatory fields. - People must be motivated, training alone is not enough. We tend to go after people who don't enter data correctly. Consider instead rewarding (not with $$$ but with public recognition) who DO enter complete and correct data. This means regular data quality audits. Get people who input date to do the audits; you may be surprised by the change in their attitudes! - Regards, V.Narayan (Vee) Lead Author, 100 Years of Maintenance: Practical Lessons from Three Lifetimes, Industrial Press.NY ISBN-13: 978-0831133238 Author, Effective Maintenance Management: Risk and Reliability Strategies for Optimizing Performance, 2004, Industrial Press NY ISBN-13: 978-0831131784 |
||||
|
There some guidelines to gather quality data in ISO 14224.
The CMMS and reliability expert should specify the minimum mandatory fields which are useful for subsequent analyses. He should also be called to enter the right data if users got stuck or to add extra codes if necessary. Users should be properly trained on the failure codes especially technicians and engineers. Of course, the failure codes must be well prepared, preferably by experts of specific equipment to know their expected failures, not by ordinary engineers or IT engineers. When using "Other" code, should be able enter free text in the proper CMMS to elaborate. This message has been edited. Last edited by: Josh, |
||||
|
Josh,
I think you have not followed my explanation. In my view, the biggest obstacle to good data quality is lack of motivation, not inadequate training or procedures. People think and behave in complicated ways, no two people react in the same way to a given stimulus. Influencing their attitudes and behaviors is not an easy task. Motivation. as you well know, comes from within, not by somebody else telling us what or how to do something. Training and Procedures do not succeed by themselves; the people who enter data must have the urge to do it correctly. While we often do a lot to de-motivate people , there is not that much we can do to motivate them. I have tried to list the few things we can actually do that can help us. You mentioned ISO 14224; the main focus in the ISO is to define the taxonomies and the system boundaries for data collection. Such guidelines that are available are procedural. The best procedures in the world will not help motivate people. By the way, I was on the OREDA Steering Committee for several years; the ISO is derived ebtirely from the OREDA Guidelines. Regards, V.Narayan (Vee) Lead Author, 100 Years of Maintenance: Practical Lessons from Three Lifetimes, Industrial Press.NY ISBN-13: 978-0831133238 Author, Effective Maintenance Management: Risk and Reliability Strategies for Optimizing Performance, 2004, Industrial Press NY ISBN-13: 978-0831131784 |
||||
|
I agree with you views, Vee. I just mentioned ISO14224 to inform others about the guidelines in there and to build on your comment about poor data quality.
I hope the CMMS engineer, Maintenance planning engineer or Maintenance Manager can explain why we collect certain data to motivate people to enter data correctly and to show how those data can be used meningfully. Simple analyses can be made available immediately so that the data enterers can see the final results and possibly their KPIs. You are right we need to motivate people to collect data and it would be nice to have a boss who can appreciate the importance of data colection and anayses like your goodself. With motivation available, those who are aware of Weibull technigues will surely apply them meaningfully. |
||||
|
Josh, Sam,
You are absolutely on the ball when you say
Sam, we can now get back to the use of Weibull. When data is of adequate quality, it is worth doing Weibull Analysis, because we can predict the time of failure probabilistically. By doing so, we can evaluate the risk of 'postponing the maintenance work' and thus extract a higher useful life from the component when the risks are relatively low. Also when we get a shape factor between 0.9 and 1.1, we can investigate the possibity of using CBM as the preferred strategy. If it is below 0.9, we are probably having premature failures, so RCA is justified. Weibull analysis gives us knowledge about failure distributions, and with that knowledge, we can decide the timing of maintenance logically and mathematically. Regards, V.Narayan (Vee) Lead Author, 100 Years of Maintenance: Practical Lessons from Three Lifetimes, Industrial Press.NY ISBN-13: 978-0831133238 Author, Effective Maintenance Management: Risk and Reliability Strategies for Optimizing Performance, 2004, Industrial Press NY ISBN-13: 978-0831131784 |
||||
|
Vee,
All very good - just how you would expect to see it in a text book. Great stuff - and I enjoyed your account of the time when you had lots of data but it was not much use - been there a few times. Sam ... you are not the first to ask for examples (there have been at least two other requests that fell flat). Once again we find "old mother hubbard who went to the cupboard". There are valid case studies and people have told me about applying Weibull to figure out the best time to replace boiler tubes. No doubt about solving the problem when there is one dominant failure mode and good data. Unfortunately the Weibull theory turns sour when the data is poor as Vee points out - and few if any organisation have decent data, but it also goes sour when there are more than two failure modes - which is commonly the case. Weibull is very clever maths indeed, but lets go back and figure what it really is... It is a formula that has variables and the shape of the curve changes according to the value of the variables selected or imposed by correlation. People who have data try to match their data with a distribution curve built around the Weibull formula. The thing most people forget is the first assumption is that the data actually behaves like the Weibull curve - what happens if it is some kind of other distribution - it could be the Hara-kiri distribution. Then the data will not be a perfect fit so the confidence could be very low - but we soldier on because lots there is lots of hype about Weibull magic and there is lots of "fun to use software" out there. It is much more fun for a junior engineer to get on the computer and produce some wonderful models than get what data is available, put it into a pareto chart and head down to the workshop and talk to some people that may be able to figure what it all means. So there you go Sam - glad you asked. I too - am waiting for a deluge of examples. The hype Weibull gets, and the amount of money people are pouring into software and modelling must mean there are plenty about. PS - we did get close once http://maintenanceforums.com/eve/forums/a/tpc/f/209103451/m/4511042592/p/1 Someone posted a problem to be solved to illustrate how Weibull works - asked for some suggested answers - but he disappeared and has not been heard from since. Have a good day Steve |
||||
|
Steve,
Good to hear from you. I have some responses for your posers, my comments are in italics.
But we need good data, run-hour or calendar dates, failure modes info, whether PM or CM activity basic repair/replacement history etc. I like Josh's ideas on how to improve data quality, and maybe one day, we will discuss the nuances of Weibull instead. Regards, V.Narayan (Vee) Lead Author, 100 Years of Maintenance: Practical Lessons from Three Lifetimes, Industrial Press.NY ISBN-13: 978-0831133238 Author, Effective Maintenance Management: Risk and Reliability Strategies for Optimizing Performance, 2004, Industrial Press NY ISBN-13: 978-0831131784 |
||||
|
With predictive technologies in place: PdM Vibration, IR, Oil utilizing CBM and RCA --- the solution is data from these PdM technologies, so why take a step backward and do three times the work to obtain what you already have. I'm with Steve on this one: good for the junior engineer that won't develop an interface with real world maintenance of getting the job done for some time to come. Creating another piece of software that incorporates 'busy' and looks good in charts to upper management may not improve the machinery or help the maintenance program. Sound like sour grapes? I don't think it is; I think time for training and getting the nuts and bolts turned to improve how our data looks is a KPI. If we are not realizing an estimated maximum life expectancy from our machinery nor experiencing smooth operation, then we need to address problems and not become one ourselves - like a pastor of a church once told me; don't be so heavenly minded that you're of no earthly good. We can install five software programs and add twenty engineers but only reduce ROI in the long run and still have the problem. The wrenches have to be pulled. Vee, I enjoyed your comments and agree. No doubt there's a fit but ultimately the report may not get the pump installed correctly. As you said, this may lead you to CBM or RCA but if you already have those tools do you need a program to tell you which tool you should use? Cordially, Sam |
||||
|
Sam,
You said
The data we need for reliability analysis is not the same as the ones we get from predictive technologies. I am sure you will agree that we cannot compute the MTBF from a set of vibration readings or thermographic images. We need start and stop dates, cause of failure or if the component was replaced premauturely (on an opportunity basis) to do reliability analysis. Contrary to a commonly held belief, PdM is NOT universally applicable; it is not a silver bullet. In determining maintenance strategies CBM is useful mainly when failures are NOT age-related, or follow an exponential distribution (with a Weibull Shape Factor of 1 - which I expand to include values from 0.9 to 1.1 somewhat arbitrarily. For age-related failures, caused typically by fatigue/wear/corrosion/erosion etc., so called contact damage phenomena, age based PMs are generally the best strategy. For non age-related failures, CBM is usually the best strategy. For early failures (Weibull shape factors less than 1), there is often a quality problem, though sometimes it is also caused by the bedding-in process. I suggested RCA for these mainly to suss out the cause of quality issues. Understanding the failure distributions is an essential step in determining strategies. For this we need the appropriate failure data as stated at the beginning of this note. In my effort to bring clarity, I hope I am not causing confusion. If so, my apologies. This message has been edited. Last edited by: Vee, Regards, V.Narayan (Vee) Lead Author, 100 Years of Maintenance: Practical Lessons from Three Lifetimes, Industrial Press.NY ISBN-13: 978-0831133238 Author, Effective Maintenance Management: Risk and Reliability Strategies for Optimizing Performance, 2004, Industrial Press NY ISBN-13: 978-0831131784 |
||||
|
Agree - but I write these things because too many engineers (IMHO) dont ever understand what Weibull is and what it is not |
||||
|
Steve,
Let me try to answer your points. Quote Vee - how did you use the parameters? Of themselves parameters do not produce shareholder value.Unquote We used it e.g., in risk analysis - QRA, and in RCM studies, Mathematical Modelling etc. We produced a database and put it on the server, so that Maintenance, Projects, Safety and otehr interested parties had access. We were able to support our case for extending test intervals with the Regulator. I think the Shareholder got reasonable value. By the way, I do not support any work that does not bring business benefits. Quote ... most people forget is the first assumption is that the data actually behaves like the Weibull curve - what happens if it is some kind of other distribution - it could be the Hara-kiri distribution. Unquote I believe that most people are justified in this instance, whether their belief is founded on knowledge or heresay. Quote Good stuff. But if you have more than two failure mechanism, Unquote As far as possible we should work with single failure modes. But you are right, we do get mixed modes. What generally happens when we mix modes is that the shape factor gets lowered to approach 1, i.e., an exponential distribution. Quote ..then you have more that one of each parameter so it becomes critical to understand at what level the analysis is done and what decisions flow from the use of the outcomes Unquote Trying to do fancy footwork without knowing to dance is a common disease with some of us. Gaining an understanding of what is really going on is vital. This is one reason I have difficulty when people reject techniques like Weibull without a gaining a good understanding. For them it is then much better to stick to Pareto and Bad Actor analysis. Most of our problems with ANY analyticcal technique lies with data quality. Weibull is often blamed because our data is bad, not because it is a bad technique. Regards, V.Narayan (Vee) Lead Author, 100 Years of Maintenance: Practical Lessons from Three Lifetimes, Industrial Press.NY ISBN-13: 978-0831133238 Author, Effective Maintenance Management: Risk and Reliability Strategies for Optimizing Performance, 2004, Industrial Press NY ISBN-13: 978-0831131784 |
||||
|
I vote best of both worlds - pareto and wrenches for the boys - Weibull for the regulatory authorities. The two don't mix.
Great thread! Mike. |
||||
|
| Previous Topic | Next Topic | powered by eve community |
 | Please Wait. Your request is being processed... |
|
|||
| Site | Join or Manage Your Profile | ||
| Servlet | eve2da019 | ||
| Version | 1.2.22 build 9062 | ||
| Module | Forums 4.0.3 | ||
| Stylesheet | "Eve" | ||
| Wordlet Set | "Default Wordlet Set" | ||
|
Contact Us | Sponsored by the Association for Maintenance Professionals | Privacy Statement | Terms of Service
Copyright © 2004-2008 NetexpressUSA Inc. All rights reserved.