Manual vs automated fault detection by vibration

[AX3L]

Hi everyone!

As a newly graduated control system and automation engineer I got a bit of a shock when I discovered that the automatic fault detection at the company I work at is non existing. Sure, they measure vibrations once in a while manually to detect bearing fault and imbalances but considering how cheap and powerful microchips and PLCs are today I can't see why they don't do it automatically all the time. And since I've not gotten any good answers I'm curious about how it looks at the industry today. Is automated fault detection just something they teach in school but that's to expensive to implement in real life today or is just my company 10 years behind? It's a large truck manufacturer by the way, so there are a lot of turning, milling and drilling going on and it's not a small 10 machine business.

Some of your inputs and thoughts about this subject would be deeply appreciated. Thanks in advance!

Axel

 

[Tmoose]

Equipment condition can and often should stand alone.
Some processes, and even power requirements could not care less about a ball/roller bearing that is softly crying itself to sleep due to lubricant degradation. But, if I don't sample vibration often enough, or ignore the signs that are pretty well detected by a little more sophisticated vibration analysis I will miss the golden opportunity to re-lubricate and in a few weeks or minutes (dn related, among other factors) the bearing will un-necessarily be turned into smoking junk requiring extensive rework to the shaft and housing AND replacement bearings with a 12 week delivery.

Temperature measurements made on the outer race can be pretty good for detecting bearing issues too, as can oil analysis looking for both wear debris and dissolved metals, especially for gears and plane/plain/journal bearings.

The equipment sellers and service providers that refer to their wares as "Predictive Maintenance" "Reliability based maintenance" or "condition monitoring" are telling the truth.

 

[GregLocock]

Hang on, nobody is saying that condition monitoring is a waste of time (I think) we're just saying that the cost of retrofitting permanent instrumentation for each machine may not be justified by a CBA.

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[spciesla]

"...plane/plain/journal bearings"

Don't forget sleeve bearing.

 

[Tmoose]

Hi Greg,

You said - "In 35 years of using lathes and milling machines I've never had one of those faults come up. "
I was thinking you were discounting the "bearing faults" AX3L more than hinted at, of which there are several ( discrete spalling on inner race, discrete spalling on outer race, discrete spalling on balls, general roughness, plus a few more.) Granted they don't happen every day, and if lubrication is reasonably good then catastrophic failure is generally not imminent, so a periodic walk around program would catch many if not most, but lube related problems can develop thermally reduced preload and then progress quickly (just seconds until the smoke gets out at 15,000 rpm).

IRstuff said - ' My point is that you need to have a rather extensive database of failures and symptoms to even begin to decide whether to perform a repair. This is further compounded by the fact that many symptoms don't necessarily lead to immediate failure."
...." For example, our company did a life test on a bunch of pumps and used a 10% increase in power draw as an indicator of a possible fail. One pump did indeed fail soon after that, but another one ran for an additional 10,000 hrs and never actually failed its other performance requirements in the duration of the test."
My 27 June reply was intended to suggest that -
1 - some fairly detailed generic vibration limits exist that could be used right out of the box as a starting point for effective condition monitoring. Yes, calling for overhaul at the first indication of BPFO may be a little premature.
2 - I'd expect looking for increased power as an indicator of a developing problem would likely miss a LOT of problems.

 

[AX3L]

IRstuff: I'm not sure what you mean by "process measurements", but I guess you mean to measure the product itself and that's of course something we do on a regular basis. The thing is though, since much of the measurements is made manually we usually run say 50 pieces between measurements (sometimes even more, and of course not everything is measured every 50 pieces). And again, since many faults like worn out or damaged cutting tool and small imbalances could be detected by vibration analysis, my hope is that we can detect the faults early on and do something about it instantly. This will save not only time but obviously money because throwing away an almost finished piece could be very expensive indeed.

Tmoose: That's exactly what I'm talking about. Just to give an example a couple of days ago we had to order a ball screw with plane and pay a couple of thousand bucks just to get it here quickly, way more than what the screw itself was. And this is something we have to do on a more or less regular basis. Using automated fault detection, we probably could have detected the problem at least a day or two earlier and sent it by regular express delivery or even standard delivery instead (ball screw faults is something you can, at least in theory, detect using above mentioned analysis).

Greg: Speaking of the investment cost, it should be said that most (if not all) of our machines already have a PLC that have more than enough power to make some vibration analysis as well. So the only investment, besides all the man hour to fit such a system, is the sensors (that even sometimes are fitted to make the manual vibration analysis easier) and some chip that read the sensors and send them to the PLC.

spciesla: Well, the list of possible faults to detect could be made very very long. I'm not sure how many extra sensors would be needed to do this though, and have therefore concentrated on the bearings at first. Of course I'm open to detect all kinds of fault though, so if anyone have any additional information of those I would love to hear them as well.

Tmoose (last post): Thanks for clearing that out! The win in reduced predictive maintenance is just as important (if not more) to me as the decreased risk for immediate failure. Besides, knowing what parts to change in the predictive maintenance would be very useful indeed. The procedure now is usually "check ... for faults", and if any is detected we have to order that part and do a second stop of the machine to actually change it. But I don't know, this theory might fail because of the limits in reliability of this analysis and we therefore still would have to check it manually with given intervals?



Thanks again for all great answers, I've learned a lot and have been given many great aspects I hadn't thought of already!
Axel

 

[ivymike]

when attended machines, or machines that are near attendants, begin to fail, their sound signature changes. That's basically what you can detect with accelerometers and such, but it happens that people can detect sounds too, and will report a change in funkiness of any given machine

Let me go back to the comment above, and related it to an experience of mine. As a much younger engineer, I got a call from the factory floor one Friday (always Friday, isn't it?) and they reported a faulty piston. Somehow this one had slipped through all kinds of QC procedures and made it into production. How did they find it? The test cell operator thought the engine sounded funny, so he took it offline and they tore it down to investigate. The piston was out of tolerance - a lot - but I'll be damned if I could hear the difference. This guy heard it from in front of a bank of test cells which all had running engines, and was confident enough to pull it out of production.

...So, now we're worried that there might be more. We stop production. We get the piston supplier on the phone and "persuade" him to come help us figure out how to find engines with bad pistons in them - given that we know that they sound different. Unfortunately, with all manner of instrumentation on the engine we were not able to devise a repeatable method of detecting this defect - even knowing that we had it. "It sounds funny" was a perfect explanation for the guy who runs the test cells every day, but just try to convince a computer of it. In the end we were able to trace the defect back to a process exception at the piston plant and contain the problem within a group of about a hundred engines, which we then had to tear down and inspect (beginning Sunday evening, in another state).

Would automatic fault detection have helped? I doubt it...we certainly gave it a shot after we knew there was a problem. The manual version worked well enough though.

 

[AX3L]

ivymike: Of course you're absolutely right about the man been the best fault detector there is. And many faults are indeed detected that way even in our company. But there are a few problems, many of them being related to the mentality of the operators of the machines. When the service technician speaks with the operator he often get to hear things like "well it have sounded like that for a week now, but it obviously still runs so i thought...". If we tell them they have to listen to the machines we would probably get a lot of false alarms as well, especially from the more inexperienced staff. It's a good point you make though!

About the computers ability to find such things I don't have the experience to tell (obviously, that's why I started this thread in the first place) but I think it would. With continuous monitoring one can detect even very small changes in the vibration spectra and relate them to the rotation speeds. You're probably right that just doing one analysis wouldn't tell if there was a problem, but if you can see the change over a period of time I think you at least can detect that there's something wrong.

By the way, does anyone know if using a regular 100mV/g accelerometer (measuring up to 20kHz), a inductive sensor to get the RPM and a relatively simple DAQ connected to a regular PC works "in the real world"? To me this seems like a great alternative, where you with cheap components can do advanced analysis using a MATLAB software. At least in theory, you could even communicate with the PLC of the machine with the DAQ and more importantly, since the computer have internet access it can send reports and data to the service engineers.

Thanks everyone who have posted so far and took time to read my long and probably bad written posts, I really really appreciate it!

Sincerely
Axel

 

[GregLocock]

Inductive sensor for rpm sounds a bit odd. How would you do that exactly?

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[AX3L]

Greg: What an inductive sensor do is measuring the magnetic field (and by that distance) to a metallic object. If you place it where you have say two bolts or any other "defect" to the plane surface on the spindle, you'll get two dramatical changes or pulses per revolution and combined with time you'll get the RPM. It sounds like black magic or voodoo, but using this method you can measure up to 250k rpm with as little as 0.1 accuracy for not much money at all and almost without need of maintenance.
Here's a link to a sensor of this kind I got from Google, just to give you an idea of what I'm talking about:

http://www.monarchinstrument.com/product.php?ID=41

 

[GregLocock]

No, I'm familiar with the device, as that is what we use on ABS tone wheels and crank position sensors. It's just the machine tools I see tend not to have bolts sticking out, some might even assume the designers have tried to minimise the chance of things catching in rotating shafts.

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[AX3L]

Greg: I see! Well I must admit I haven't checked it out in detail since I know others have similar solutions that work. But the three "claws" holding the piece in the chuck of the turning machine seems possible to sense in some way. Or have you already tried that?

 

[GregLocock]

That,d work in that case, but milling machines grinders etc wouldn't. There are plenty of alternatives, a blob of paint and an optical tacho being the easiest in many cases.

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376

http://eng-tips.com/market.cfm?

 

[AX3L]

Optical is what we're using now when we check manually so I guess too that's an alternative. I can't see how it will work over time with all the fluids and/or dust in the machines though. Another alterative is of course to solve this using algorithms instead. We usually run 400-500 parts of each of, and it wouldn't be impossible to let the operator input what part we're doing in the computer. Or using the software to identify it itself by analysis of the spectrum over a period of time and see what predefined "master spectrum" is most similar to what we're doing right now.

In some cases I think even the machine itself measure its rpm, and communicate with the PlC of the machine using either a digital or analogue output can't be too difficult.