I'm in the process of developing a device to monitor motor/pump/generator degradataion based on current signature analysis, and potentially vibration and temperature. Can anyone provide their input on what they didn't like about similar products, and what they'd be looking for in something that's on the drawing board now?

I'm anticipating it being used on small- to medium-sized industrial motors and pumps. Current specs have it monitoring current signatures with Hall effect sensors; embedded artificial intelligence that will estimate normal behavior and compare it to current operating modes and measurements; wireless connectivity; and form factor would be about the size of a pack of gum.

Any input would be greatly appreciated.

MJroc,

It's about performance: faults types, early warning, and measurement frequency or interval.

It's about cost: hardware, software, and installation labor.

Some low power wireless systems have a slow measurement interval of 45-minutes or longer, so transient operating conditions or fast fault modes can go undetected. Want to discuss further?

Walt
w_f_strong [at] msn {dot} com

What method of AI are you going to use?

I have a design that is wireless and battery powered. Should be able to get about a year out of the battery and does not use RF. About the size of a pack of cigarettes. 5 chs. Sigma Delta or 8 chs. at 14 bit A/D.

Mr Hatfield,

Can you provide details of your wireless system?

Walt
w_f_strong {at} msn [dot] com

Mr Hatfield,

Also interested, we have some basic rule based system.

Walt did you get my e-mail?

This system design would transmit 40 KHz encoded TWF data via IR. It utilizes existing technology from 3 different areas of which all are mature. It is all in how it is put together.

From my perspective the most important part of such a system would be some degree of credibility in the results. There has to be an experience base and the user has to be able to look at what the machine saw and see how it came to the conlcusion that it came to. Otherwise it may be tough to convince anyone to take action from the recommendations of the box.

For motors I think current signature analysis is in it's infancy. I have a hard time imagining drawing many firm conclusions about the pump condition from the current.

You are exactly right E'pete. However, there are ways to do exactly what you talk about. Ever see the movie Hunt for Red October? the basis of the sonar diagnostic system is one such system. It takes many pieces and parts (Data from various sources) to come up with a final conclusion. Even then it is just a probability factor. Several years ago we tried to bring the technology to the industrial world. A few understood some of it. Most wanted it handed to them. It is now locked up in the world of the Lockheed Martins and other forward looking companies.

I agree that credibility is crucial, but it's like a chicken-and-egg quandry--no one is interested without results, but you won't get results unless someone is interested. I've inquired about people being beta testers because it appears it would achieve both goals; the user gets their hands on something for reduced or no cost in exchange for feedback. But credibility (especially in a fault detection product) would be key.

The current state of the system (pun intended) would use embedded controllers; the algorithm very efficiently models the system on-board; board is battery-powered and/or USB-powered; at this point wireless comms are done with Bluetooth to a PC. 24-bit A/D on 10 channels (but resolution hurts the throughput, so will likely exchange lower resolution for up to 100 kHz of short-burst high-bandwidth data) COTS-components are used, but programming the embedded system customizes it.

I certainly agree about the credibility issue. I believe that is why the "Expert" systems sold some years past failed. The rules were canned. No one could view nor edit them, at least not without spending much more than what they initially invested in them.

What we did in the NeuroWatch system was use Tech Associates of Charlotte diagnostic guide as a starting point. Taking care to pay attention to data resolution, bin separation, windowing, and etc. We calculated the frequencies and faults that would be generated by the machinery we used as a pilot. Data collected was TWF, fft's and phase. In the case of twf post processing was done if needed.

Discrete data was extracted and fed into a Class Dependent-Elliptical Basis Function Neural Net (CD-EBF) in parallel with a boolean based rules system built with MS Access. We used archived data to do the initial training of the the NN. Initially we used the linear rules until further training was accomplished.
(not all neural nets are the same) We have data to back it up and were able to detect operational transients.

Also used were Evolutionary Programming techniques which extract the rules or why the system made the call it did. The user built the rules and identified the data patterns and named the faults. If there was any discrediting it is upon the user. The user essentially builds the algorithm. This type of NN allows overlap of faults just like in the real world.
All a NN is, is a statistical data patterning tool which provides meaningful info to the human. Not a replacement for the human.

As for the data acquisition system I spoke of it is meant for short range communictions. Like in a submarines equipment compartment. It utilizes a modified digital motor controller and encodes TWF data and transmits it off the system and goes back to sleep. It's meant to be an automomous PdM system. All other processing is done via software on a PC. Other comms methods are also available.

Take a look at what UC Berkely is doing in the area of macromotes. Take a look a Lockheed-Orincon as well.

quote:

Originally posted by MJroc:
I'm in the process of developing a device to monitor motor/pump/generator degradataion based on current signature analysis, and potentially vibration and temperature. Can anyone provide their input on what they didn't like about similar products, and what they'd be looking for in something that's on the drawing board now?

I'm anticipating it being used on small- to medium-sized industrial motors and pumps. Current specs have it monitoring current signatures with Hall effect sensors; embedded artificial intelligence that will estimate normal behavior and compare it to current operating modes and measurements; wireless connectivity; and form factor would be about the size of a pack of gum.

Any input would be greatly appreciated.

Baker Instrument has an excellent tool for diagnosing mechanical issues from the MCC by just measuring the voltage and current quantities. The Explorer does have the ability to perform current signature analysis but the best tool to evaluate the load and mechanical issues is the Torque. The torque is a great way to separate between mechanical and electrical problems because the load is what requests more or less torque from the motor. If something changes in the torque the load is most likely the culprit. The torque also has a much stronger signal to noise ratio allowing the fault frequencies to be detected earlier than with the torque.

Electricpete has hit the nail right on the head. Credibility is the key. Case studies are what are needed to convince people the test equipment can find these problems. I will attach a couple for all to review.

Sorry for a late response. Hope this will help to demonstrate how the Torque can diagnose mechanical issues.

Predictive_Maintenance_of_Mechanical_Failures_using_Electrical_Measurements_for_Instantaneous_Torque._A_Modern_Approach..pdf (842 Kb, 36 downloads)

quote:

Originally posted by MJroc:
I'm in the process of developing a device to monitor motor/pump/generator degradataion based on current signature analysis, and potentially vibration and temperature. Can anyone provide their input on what they didn't like about similar products, and what they'd be looking for in something that's on the drawing board now?

I'm anticipating it being used on small- to medium-sized industrial motors and pumps. Current specs have it monitoring current signatures with Hall effect sensors; embedded artificial intelligence that will estimate normal behavior and compare it to current operating modes and measurements; wireless connectivity; and form factor would be about the size of a pack of gum.

Any input would be greatly appreciated.

Attached is another case study utilizing the Torque to find mechanical issues.

Comparison_of_Duct-Mounted_Vibration_and_Instantaneous_Airgap_Torque_Signals_for_Predictive_Maintenance_of_Vane_Axial_Fans.pdf (584 Kb, 31 downloads)

That's a good article. Where did it come from? (are there more articles available from that source?)

There is an extra variable introduced during current measurement which may not be present during vibration measurements - variability of the main current with load. I would think that the exciting current is modulated by whatever vibration is present and is constant with respect to load. The load current would be independent of the defect. So at high load the signal you are looking for is buried deeper in the grass than at low load. For an axial fan such as in the case, highest load comes when you provide a restriction to airflow. From the picture and a casual converation I had at a conference with someone peripherally familiar with this experiment described in the article, I am under the impression the fan was operated with no backpressure which corresponds to minimum load (again for axial flow fan... opposite for radial fan). So what was detectable here may not have been during normal plant operation where the is a backpressure on the fan.

Pete,

Even though I'm not from Missouri, the show me state, I prefer to prove it to myself before believing or trusting anything new. I'm new to the Baker torque spectrum analysis but it looks like it is naturally demodulated like they say. One of the many neat things about this Explorer online tester is the fact that it gives you all three spectrums, volts, current and torque! As I recall, the torque spectrum goes up to 23K hz. select any part of that spectrum and zoom in/out. For better resolution anywhere in that spectrum, simply select a different span!

I do have one case study in progress showing stator mechanical in the torque and current spectrum. It's so much nicer not having to add and subtract 60hz from everything.

RS X #SS = SM peak in torque

RS X #SS = Fref +/- 60hz in current

The neat thing about comparing the two spectrums is the fact that the current peaks do not show in the torque spectrum and the torque peak is not shown in the current spectrum.

By golly, I think I've bought in to this new technology

Thanks Jim. We haven't bought any current signature equipment yet. We do use a current clamp-on with our Entek datapak and look for pole-pass sidebands around 60hz... but that's about as far as we go. I have been watching and listening for awhile now a lot of new developments. Sooner or later it will be time to take the plunge. Sounds like you are satisfied with the Baker equipment.

One thing that strikes me is that there is quite a bit more effort involved in setting up to monitor all 3 phases current and all 3 voltages, rather than just one phase current as we do now. For our 4kv and 13.2kv motors (which covers most of our critical motors), we simply open the switchgear door and clamp on to the panel ammeter wiring which is fed from the ct. I think we could get the other three currents pretty easily (clamp on to a relay input) but I think there is quite a bit more effort and risk when hooking up voltage probes.

Can you get good information just looking at the current?

Hi Pete,

Yes, rotor bar pole pass around line frequency works very well in the current spectrum, with the three online testers I've used, pretty straight forward, no need to know rotor bar count, just speed.

Baker, Framatome and PdMA all offer plug in devices for quick and easy monitoring for online analysis. This eliminates the need for any heavy duty arc flash personal protective equipment and again, less hastle, just plug in and test. All three can use the existing MCC low voltage MCC panel metering circuits as their source, like you are doing now. If metering circuits aren't available, then the CT's and PT's need to go to the actual circuit wiring. Obviously sizing is the important thing here.

I'm not sure if Baker is the only one that recognizes the motor circuit when you plug into the receptacle or not. I'm sure they all will sooner or later, again eliminating some more hastle as the operator doesn't need to open the folder. Less time to setup, more time to test more motor circuits.

Current works well for rotor bar/shorting ring and cast aluminum rotor void problems, however if you want to look at other frequencies, you really need the voltage spectrum to see if your suspected peak is being generated there, rather than some load related problem. With the torque spectrum in the Baker Explorer, cross checking the voltage spectrum is not necessary because the torque spectrum is naturally demodulated and not affected by the voltage spectrum.

Yes it takes more effort to get a complete picture of your motor circuit health but this is needed when looking at power quality, power circuit problems, motor related problems, as well as things connected to the motor shaft. The Baker torque spectrum is also capable of finding bearing frequencys (Framatome in current), coupling problems, imbalances and other mechanical created problems on the other end of the motor shaft. This works great for those machines where the vibration guys can't get to the equipment, like a submersible pump or perhaps maybe due to some added machine guarding or safety rules preventing them from riding machinery while its in motion.

Yes, you can get good information in a current spectrum, but like I said before (and I'm sure Motordoc will agree as he is the one that set me straight on this one), one needs to confirm that the same peak does not show in the voltage spectrum. However, Baker and Framatome are the only ones that give you the voltage spectrum to look at in hz.

Good questions!


Jim