Cost-effective preventive maintenance scheduling

[alexduque89]

Dear all,

Maybe you can help me with this:

There is a factory producing packages thorugh 40 IDENTICAL 3D laster cutter machines. At the moment the management of the factory is experiencing problems with the MAINTENANCE of these machines, because the maintenance is highly reactive (corrective mostly) and they want to move to a more proactive approach. The problem is that at the moment they are not being able to perform enough preventive maintenance tasks for the machines because the machines fail constantly and this workload makes impossible for the technicians to perform scheduled preventive maintenance tasks. The company aims to perform on PM a day at least (i.e. all the machines would be covered within 40 days), however they rarely manage to perform enough PM tasks. Thus, as you can imagine, and taking into account that the machines are pretty old (wear-out phase), the failure rate is high. My task is to improve this scenario, maybe suggesting the hiring of more technicians or maybe suggesting a NEW more COST-EFFECTIVE MAINTENANCE POLICY (basically a new PM schedule).

My question is: is there a way of calculating the effect of a PM task in the failure rate? This is: suppose that I suggest that they should hire a new technician (for example) and perform 4 PM per day so that all the machines are covered within 40 days. As I can imagine, since they would be doing more PM the failure rate is expected to fall... But, is it possible to quantify (or at least estimate) this expected reduction in the failure rate?

I am asking this because the method I am thinking to use is simulation. I want to build a model of the current maintenance policy used in the factory which would be validated comparing it with the actual real-life scenario (comparing obtained availability, etc.). Later, I would modify this model in order to try different scenarios and PM schedules. However I can imagine that if, say, I increase the number of PM the failure rates should be expected to fall, so... how could I model this? (or estimate).

Thank you.

 

[MikeHalloran]

I'm not convinced that modeling what you've been doing will teach you enough about what's going on, soon enough to make a difference. If you're talking about a couple of days work, fine. If you're talking about months of delay before you can take action, some action that you won't like may take place first.

Instead:
Buy a service contract from the OEM for at least a few of the machines.
Borrow or lease a new machine from a different manufacturer.
Make sure that the OEM's techs happen to see it.
Either they will manage to make your existing machines work a lot better, or their sales team will find a way to cut you a good deal on new replacements.


Review the (presumably long) history of unplanned service incidents, in enough detail to see any correlation with the PMs that you have managed to squeeze in.

If you are PMing parts that don't fail much anyway, or give some detectable warning of impending failure, maybe the PM is a waste of time.






Mike Halloran
Pembroke Pines, FL, USA

 

[IRstuff]

Bear in mind that some PM processes involve a slight bit of stress testing, which, in a previous company, would clobber our ion implanter for about two weeks after the PM.

TTFN
faq731-376

Need help writing a question or understanding a reply? forum1529

 

[MikeHalloran]

One more data point about PMs that I had forgotten in the ensuing decades:

Our software development process was critically dependent on half a dozen expensive Microcomputer Development Systems. Monthly or so, as contracted, the mfgr's service guy would show up, remove all the circuit boards from the racks, and clean the dust off them with a horsehair draftsman's brush. The brush did indeed remove the little dust that had accumulated. It also dislodged the whiskers growing out of the tin-plated IC sockets then in use. Some of the whiskers fell into places where they shorted signal traces to each other or to a rail. So every PM was inevitably followed by a day of troubleshooting and another day or two for delivery of replacement parts for at least one of the workstations.

At the same place, the engineers assembled their own homebrew computers from the same manufacturer's parts, nominally supplied as spares for our end products. Those computers never got a PM, and never gave trouble.

Like I said, look for a correlation.



Mike Halloran
Pembroke Pines, FL, USA

 

[IRstuff]

I think you need to do some ground work to analyze what the failures actually are, and whether preventative maintenance is even plausible. Wearout failures could potentially be detectable and repairable before they fail, but not random failures. While there has been much discussion about prognostic replacement of parts with high operating hours, the variability of failure times means that pre-emptive replacement will cost substantially more in replacement parts cost, due to the fact that you will most likely wind up replacing boards well before they can fail, which means losing lots of operational life. If there's financial impact to product due to board fails, then you need to determine where the statistical break-even point is.

TTFN
faq731-376

Need help writing a question or understanding a reply? forum1529

 

[alexduque89]

Thank you all for your replies.

Now I am a bit confused, I have to confess. I am also wondering whether the management of the company is aware of the characteristics of each of the failure modes for the machines (whether they now if those are random, wear-out, etc).

Maybe it could be interesting to have a look at each of the failure modes, classify them in function of their characteristics and see if the PM procedures followed in the company make sense (if they are making PM in the right elements and not in those classified as random failing elements) or some of them are just a waste of time...

At this point, could you suggest me a method to classify the different failure modes and to determine whether those are random failing, time-based failing elements, etc? Maybe I could perform a Weibull analysis for each of the failure modes (or elements?) to determine the Beta factor... does this make sense? How is this Weibull analysis performed? this is: is this Weibull analysis going to tell me whether the ELEMENT is a radom/time-based failing element or this analysis only applies to specific FAILURE MODES for a specific element? I don't know if you catch what I am trying to say at this point.


Thank you for your replies again.

 

[MikeHalloran]

I think you should PUSH yourself away from your computer, and spend at least a week in the shop, working alongside, and especially LISTENing to, your maintenance techs.

Take notes.
Take photos.
Make videos of the operations they do.

The nominal objective of that activity is to document what the techs do and how they do it, so as to promote standardization, facilitate training of new hands, and perhaps discover nonproductive or errant techniques.

... but really, you want to listen to them bitching about the stupid design features in the product and in the tools, the silly parts of the procedures they are obligated to follow, and the stuff that breaks and shouldn't.

If you can gain the techs' trust, they may allow you to copy their personal notes. ... which will probably require some explanation, so you need to also have them explain everything step by step in their own words. Don't argue, just listen (and, better, record) what is said, and evaluate it later. Their impressions may not be supportable from an engineering perspective, but they are never wrong. Later, back at your desk, you can come to an understanding of the physics behind what the techs observe.



Mike Halloran
Pembroke Pines, FL, USA

 

[IRstuff]

If you have sufficient and accurate data, a Weibull analysis could identify the shape factor, which determines whether it's random or wearout

TTFN
faq731-376

Need help writing a question or understanding a reply? forum1529