ferrite bead for surge suppression of power circuits 1

[electricpete]

Just thinking out loud....

Ferrite beads apparently filter surges based on frequency-selective behavior:

http://en.wikipedia.org/wiki/Ferrite_bead

Could such a device be used as surge protector for power devices like motors?

It would have the advantage of introducing no additional failure modes (unlike capacitors or MOV's) which physically connect to the power system)

This approach is not used as far as I know. Is there some technical challenge or is it a cost challenge? Any thoughts?

=====================================
(2B)+(2B)' ?

 

[waross]

I don't think that beads are very effective at lower frequencies or higher currents. As many CTs are single turn transformers a bead creates a single turn inductive reactor. You may get orders of magnitude more reactance by running your phase conductors through individual 10 foot lengths of ferrous conduit.
I have a very old text book that shows a "wire less" reactor. This was a core about six inches square and six or eight inches high that was mounted on the side of a transformer. One of the secondary leads was run through the center hole of the core. The purpose was to add inductive reactance to a transformer so as to match impedances for parallel operation. This is the same effect as a ferrite bead.
As you know, the inductive reactance of an inductor is proportional to the frequency, so ferrite beads are more effective at higher frequencies.

Bill
--------------------
"Why not the best?"
Jimmy Carter

 

[electricpete]

Good points.

Yes, it's basically an inductor which is naturally a low-pass filter in lumped-element design. And as I understand it, the ferrite material keeps high permeability all the way up to high frequency range and may even have a permeability peak up at high frequency which would increase the filtering of high frequencies.

As you say, loads like motor probably have much lower impedances than electronics, so maybe the device needs higher inductance for power systems to be effective that it does for electronics... maybe that's the reason it's not used for power systems.

It doesn't necessarily need to be large enough to create an large inductive voltage drop, but it needs to at least create a sudden increase in surge impedance to reflect the wave back to the source.

=====================================
(2B)+(2B)' ?

 

[Skogsgurra]

Normal ferrite beads are exclusively used for HF suppression.

If they shall be effective for power frequencies, they need to be gigantic. The resulting impedance can be calculated from the number of turns, the permeability and core dimensions. Saturation limits the flux density and thus also current range.

Large ferrite 'beads' are used for common-mode current suppression. There is a short description in the 'EDM crash course' (google) and a recording showing what happens when a large core is subject to a capacitor discharge is shown in the attachment.

This core saturates at around 2 T, a lot higher than ferrites, which saturate well below 1 T - usually 400 - 500 mT.

Even then, the time during which the core absorbs voltage is counted in microseconds. For 50/60 Hz power applications, one needs milliseconds. That's why we do not see ferrites in low frequency power applications.

Gunnar Englund

http://www.gke.org

--------------------------------------
100 % recycled posting: Electrons, ideas, finger-tips have been used over and over again...

 

[electricpete]

Thanks for the comments.

Let's look at saturation:

For a toroid-like core core, the magnetic circuit gives us:
Phi = N*I * mu*A / MeanLength

B = Phi/A = N*I * mu / MeanLength

(area is not important to flux density, but higher area gives higher flux which gives more induced voltage).

Using values: N = 1 turn, muRelative = 1000, mu = muRelative*mu0 = 1000* 4*piE-7, Diameter = 1 foot = 1/3 meter, MeanLength = circumference = pi*D = 1 meter, current = 100Arms = 141A peak, we have:

B (peak) = N*I * mu / length = 1 * 141A * 1000 * (4*piE-7 H/meter) / 1 meter = 0.17 T
(if I did my math right)

If current tripled from 100A to 300A, I guess you'd hit the 500 milliT you mentioned for these paramters. Perhaps there are types of ferrite which give even lower relative permeability than 1000 at power frequency and higher saturation levels ( ?).

If saturation is a big concern, what about a single ferrite toroid to encompass all 3 phase conductors? (I imagine most of the surges of concern represent voltage to ground and would return through ground capacitance rather than through the another lead.)

I did not understand what you meant by "the time during which the core absorbs voltage is counted in microseconds. For 50/60 Hz power applications, one needs milliseconds". And I didn't understand the relevance of your attachment or your EDM crash course to any points being discussed.

Even if it turns out practical to build it as 1 turn without saturating at power frequencies, there is the other half of the question: how effective would we expect it to be ?

=====================================
(2B)+(2B)' ?

 

[waross]

Say Pete; Just for fun, can you calculate and compare with a couple of other low value impedances?
One would be a length of 1/2 inch rigid conduit.
I remeber years ago some test results involving currents in ground conductors that were mechanically protected by a length of 1/2 inch diameter rigid conduit. The test results showed that the conduit choked the current of at about 100 amps. I believe that the testing did not supply enough current to saturate the conduit.
Another interesting comparison may be about 10 turns of about 1 inch or two inches diameter.
As this is an air core and not subject to saturation it may be comparatively more effective at higher currents.
I worked for awhile in an area where the urban legend was that service drop conductors MUST be coiled. Each service conductor, just before entering the weather-head, was wound into a 10 turn coil using the handle of a screwdriver as a form. I am sure that the coils mitigated surges from lightning strikes but I have no idea how effective the practice is.

Bill
--------------------
"Why not the best?"
Jimmy Carter

 

[electricpete]

Thanks Bill and Gunnar – those are good thoughts to help focus my thinking.

I was thinking that it was not really the total inductance that was important, but the change in surge impedance, since dramatically higher surge impedance would cause almost complete reflection of the incoming wave.

But now I realize there is one thing I didn't think about... that reflection phenomenon applies at the junction of two transmission lines. To act like a transmission line, the "ferrite bead" would have to be something on the order of a wavelength or more? (does anyone know what multiple or fraction of a wavelength is required?).

The rise times of interest are typically stated on the order of 0.1 microsecond. The frequency content associated with that rise time would depend on the shape of the rise. The smoothest rise (lowest frequency) possible would be half sinusoid period over 0.1 seconds which would be a full period at 0.2 usec, or a frequency of 5 Mhz. Most likely a lot higher. For a frequency of 5Mhz, the wavelength would be Lambda = c/f = 3E8(**) / 5E6 = 60 meter(!). (** Actually c might be a little lower due to effective relative permittivity and/or permeability greater than 1, giving slightly shorter wavelength). Higher frequency will reduce that but I don't know how much higher the actual frequency content is.

That suggests something like a coil as Bill mentioned might work better to extend the length to the required fraction/multiple of a wavelength, but then the turn-to-turn impedance is a lot lower... independent of presence of any high-mu metal. For air core again I think the wave would propogate turn-to-turn rather than linking all turns at once. But it's something to think about.

Also worthwhile to note (and probably obvious) that there are many variables that affect the propagation of the surge to the motor. An interesting quote from IEEE62-21:

Motor supply cables individually shielded with outer jackets that effectively isolate the shields from the raceway, and the shields bonded at only one end (only at the motor end) to the metallic raceway and to the motor frame and to a low impedance ground or earthing system. (This shield bonding configuration can reduce the surge at the motor by as much as 60% compared to bonding the shields at both ends) [B23].
B23 - [B23] Dick, E. P., Gupta, B. K., Pillai, P., Narang, A., Sharma, D. K., “Practical Calculation of Switching Surges at Motor Terminals,” IEEE Transactions on Energy Conversion, Vol. 3, No. 4, Dec. 1988, pp. 864– 872, with W. G. correspondence.

Of course there are reasons we don't want to leave the supply end cable shield unbonded... would reduce magnetic shielding from the power circuits to surrounding circuits... and could result in local voltage differences in event of power system fault. And system grounding plays a role as well... complicated.

Sorry.. that's a lot of rambling. I think you guys have shown good reasons a short ferrite bead by itself probably wouldn't do much. I think I'd like to study a little more to understand how the surges propagate.

Some questions that are still on my mind:
1 – What is reasonable approach to convert the 0.1 usec rise time to frequency? I guess it will depend on shape of the rise, but perhaps there are some thumb-rules related to typical switching surges?
2 – What length of transmission line (in multiples or fractions of wavelength) is required for a cable etc to adopt transmission-line-like characteristics such as reflection?

=====================================
(2B)+(2B)' ?

 

[Compositepro]

Many ferrites are made from carbonyl iron which is pure iron powder made by a gas phase reaction from carbonyl iron. The particles are spherical and about 5 micron in diameter. A non-conductive resin binder holds the particles together to form shapes. The main feature is that the material has the magnetic properties of iron but is non-conductive so there are no eddy currents or skin effect at high frequencies. At power frequency laminated cores would work the same and be much less expensive.

 

[electricpete]

The objective is not to create a high-impedance at power frequency, (any voltage drop at power frequency is an undesired side effect). Rather, tThe objective is to create a high impedance at surge frequencies. To my way of thinking, laminated electrical core steel would not work because the eddy currents at high frequency reduce the effective permeability for those frequencies .

Some supporting info:

Table 16.5 here suggests a rapid drop in permeability with frequency for electrical steel, beginning about 10khz, even with thin 2 mil lamination thickness:

http://books.google.com/books?id=nZ...ical STEEL" PERMEABILITY VS FREQUENCY&f=false

Figure 16.2 suggests the permeability of ferrite is stable up to 3 Mhz:

http://books.google.com/books?id=nZ...&q=magnetic permeability vs frequency&f=false

Am I looking at that wrong?

=====================================
(2B)+(2B)' ?

 

[electricpete]

Just to make sure I'm comparing apples to applies, I'll refer to complex permeability as described here:

http://en.wikipedia.org/wiki/Permeability_(electromagnetism)#Complex_permeability

I gather the real part is what we typically associate with scalar permeability and inductance, the imaginary part is associated with iron losses.

Fig 7 here shows real (mu') and imaginary (mu'') parts of complex permeability for ferrite:

http://tdserver1.fnal.gov/tdlibry/TD-Notes/2007 Tech Notes/TD-07-025.pdf

The real part is stable again up to 2-3Mhz, the imaginary part increases with peak around 10Mhz.

Fig 6 here shows carbon steel real and imaginary parts of permeability.

http://home.eng.iastate.edu/~nbowler/pdf final versions/conferences/QNDE2005Bowler.pdf

At 0.1 hz, the real part is 100 and decreasing, imaginary part is 20 and decreasing. Certainly by 1Mhz they will be far lower.

=====================================
(2B)+(2B)' ?

 

[Skogsgurra]

Pete, you have dug so deep into this now that I haven't been able to follow all of your thinking. But I can explain a little more about the EDM and common-mode part of this rather complex (no pun) subject.

The EDM crash course describes how a ferrite or amorph material core increases the common-mode parasitic currents flowing through winding capacitance to ground. The parasitic currents induce voltage in the motor shaft and that voltage creates bearing currents. Reducing the parasitic currents also reduces bearing currents quite effectively.

As you can see in the attached instruction on how to install common-mode filters, the original PWM edge is around 300 nanoseconds (green trace). The filter absorbs the first part of the PWM pulse over 2 - 3 microseconds.



Gunnar Englund

http://www.gke.org

--------------------------------------
100 % recycled posting: Electrons, ideas, finger-tips have been used over and over again...

 

[Skogsgurra]

Why, why, why?! Read "describes how a ferrite or amorph material core decreases the common-mode parasitic currents"

I was probably starting to write 'increases the impedance'...

Gunnar Englund

http://www.gke.org

--------------------------------------
100 % recycled posting: Electrons, ideas, finger-tips have been used over and over again...

 

[electricpete]

Thanks Gunnar, that is interesting stuff.

I can see the relationship that the magenta line (volt-sec) is the time-integral of the green line ("absorbed voltage").

Some questions:
1 - How was the green line (absorbed voltage) determined/measured?
2 - How do you decide how many of those toroids are required?
3 - What is the rise time for PWM vfd waveforms where these ferrite beads would be used?

=====================================
(2B)+(2B)' ?

 

[electricpete]

Further on question 1 - is the green waveform the voltage pulse downstream of the ferrite? or upstream? or a differential voltage?

=====================================
(2B)+(2B)' ?

 

[Skogsgurra]

Question 1. The voltage was measured with a single turn of wire through the core. Follow-up question: Neither. It is the voltage measured between ends of the 'probe wire'

Question 2. The cost isn't very high compared to the production loss when you stop the machine to install the toroids. So we use three of them. That usually is more than needed and, since they do not harm (heat distributed between three rings is better than having it all in one or two rings), we leave it as such when measurements across bearings say that EDM is gone.

Question 3. Usually between 200 and 300 nanoseconds 10-90 %. That is the rise-time the green trace has. All of the front edge (or almost all of it) is caught by the toroid.



Gunnar Englund

http://www.gke.org

--------------------------------------
100 % recycled posting: Electrons, ideas, finger-tips have been used over and over again...

 

[electricpete]

Question 1. The voltage was measured with a single turn of wire through the core.

Very tricky. That is a good way to measure the same voltage as induced in the power cable by the core, without having to connect to the power cables. I wouldn't have thought of that.

Question 3. [the rise time for PWM vfd waveforms is] Usually between 200 and 300 nanoseconds 10-90 %. That is the rise-time the green trace has. All of the front edge (or almost all of it) is caught by the toroid.

So the rise time that you are combating with these ferrites is very comparable to the rise time of the surges of concern for motors. (IEEE 522 uses a 0.1 microsecond = 100 nanosecond rise time for motor surge testing).

Therefore, if we took these same ferrite toroids and put them in the motor terminal box, in a configuration you show of enclosing all 3 conductors, and downstream of the shield termination, shouldn't that be an effective alternative to surge capacitor (*) for protecting the motor from these surges?

(* all a surge capacitor does is lengthen the rise time of the surges, it doesn't do any voltage clamping unless partnered with a surge arrester... which is not common practice at my plant and the plants I am familiar with)


=====================================
(2B)+(2B)' ?

 

[Skogsgurra]

Yes. It will reduce the steep wavefront between motor phases and ground. But it will not reduce the steepness of voltage between phases. That voltage front is normal mode and the toroids only reduce common mode voltage du/dt.

If you look at it in detail, you will notice that motor phase A and motor phase B (and the other combinations) cancel the inductance because they all pass through the toroid in 'anti-phase', plus-minus and minus-plus or bifilar winding.

Gunnar Englund

http://www.gke.org

--------------------------------------
100 % recycled posting: Electrons, ideas, finger-tips have been used over and over again...

 

[electricpete]

Right, good point again. I'll have to think about that some more. At this point it seems reasonable to think that the voltage can appear phase to phase at the breaker during switching, travel along the as phase to ground pulses in individually shielded cable (the configuration of our mv motors) and arrive at the toroid at the roughly same time with equal/opposite polarity, rendering the toroid insensitive. In that case we have to tackle the saturation problem for putting toroids on single conductors.

The pulse clearly does not travel all the way through the motor and return the other phase in a true differential loop (the pulse doesn't even reach the motor neutral point), but that doesn't really help anything when the pulses arrive at the toroid at the same time with equal/opposite current.

And additionally, small differences in length of cable won't make a difference... assuming 0.3 usec rise time, 2.0 relative permittivity of cable insulatioon => propagation speed = 3E8/2 => the 0.3usec pulse spans 45 meters of cable. Even if we inserted 1 meter difference in length of cable, a fraction 44/45 of the edge of the pulse would still overlap the edge of the pulse of the other phase. We could start coiling the cables as Bill suggests (maybe 0 meters in phase A, 45 meters in phase B, 90 meters in phase C) to get ourselves that 45 meter difference to ensure edges didn't overlap, but that would be cumbersome, could cause heaitng concerns, could get expensive among other things. Not worth further discussion.


=====================================
(2B)+(2B)' ?

 

[electricpete]

Correction: change speed (3E8)/2 to (3E8)/sqrt(2).. 45' to 65', 90' to 130', same conclusion.

=====================================
(2B)+(2B)' ?

 

[electricpete]

One thing I was thinking of regarding saturation of the ferrite... perhaps we could wind a coil onto the ferrite, then load that coil with a loading inductor (also wrapped around a ferrite core, but not enclosing any primary current). The original ferrite acts roughly like a CT..... with low impedaance load at power frequencies keeping flux low. At higher frequencies, the loading inductor impedance increases since we want to present high impedance on the primary for high frequencies. Just a thought... although the more complicated/expensive this thing gets, the less attractive since surge caps are by and large pretty cheap and reliable.

2 – What length of transmission line (in multiples or fractions of wavelength) is required for a cable etc to adopt transmission-line-like characteristics such as reflection?

I have been looking for answers of this question and as far as I can tell, there seems no lower limit. The "transmission line matrix method" uses arbitrarily small lengths of transmission line and assumes they still act like transmission lines.

So, I gave a try at analysing the ferrite bead using transmission line theory in attached. The results seem a little surprising... the characteristic impedance ratio (K) plays a huge role, but the length of the ferrite "sleeve" (a long bead) does not seem to make any difference... in fact by this model shorter sleeve tends to do slightly better (which conflicts with conventional circuit theory where we'd expect the longer sleeve has higher inductance and does better filtering).

I'm not sure if this is anywhere near right. Any comments?


=====================================
(2B)+(2B)' ?