I have a setup that consists of a variable frequency drive (VFD) and a PC. The two talk to each other using RS-485. (The VFD output is 3-phase, with a high frequency square wave riding on a 100 Hz sinusoid carrier wave.)
I'm running into some major VFD noise problems, and I don't know how to solve them.
The noise problem is so bad that once the VFD starts running, it kills the RS-485 signal.
The noise problem also kills our ability to use a hall current probe to sense the current through one of the 3-phase output lines.
I've tried grounding everything. I'm using shielded cables and toroids for the communication and measurement lines.
On the RS-485 end, we've tried isolated and non-isolated devices (with proper termination resistors) on the PC.
As Terry said it is probably a PWM signal that actually runs the drive and not just controls it. The ripple you see would then be the stuff that didn't get smoothed out in the output filters of the converter unit. One thing you could look at is filtering some more at the source but you may no be able or allowed to modify that, let alone get inside the converter.
I usually prefer current transformers. You can filter the output of these since you are then dealing with isolated and low level signals. Running the output of the current transformer differentially to the system where there would be another transformer may also be an option. But don't ever move the resistor away from the current transformer. Else havoc can happen should somebody disconnect the cable.
At these power levels you'd have to be a real RF expert to get it running as is. Or hire one ;-)
This situation really calls for transformer isolated transmission, not just differential receivers/drivers. The required transformers would be pretty small and in a pinch can be made by hand. Toroids, no pot cores and the like. Making your own transformers instead of relying on some "isolation device" or transformer in a can has the advantage that you know how good and symmetrical it is.
I'm using a Delta VFD-A. They no longer have this model on their website.
I don't know how fesable it will be to use screened cable. The problem is that I'm using the drive to test a motor for production, and I may be stuck with whatever cables come pre-attached to the motor.
As for the RS-485 isolation, this hasn't helped as much. I was using an isolated PCI RS-485 with built-in terminator, and I was still getting noise problems from the VFD when it was running.
Let me see what I can do, and I'll post a follow-up.
I'm definitely NOT an RF guy. I'm just a poor firmware coder who got hoodwinked into this VFD fiasco.
When you mention transformer isolation, you are suggesting that use one transformer at each end, correct? I'll try that tomorrow.
As far as current transformers... I think we tried using a current transformer and combined it with various types of LPFs. The problem we were having was that the DMM we were using would pick up the VFD noise. Noise Noise Noise! Argh!
From bitter experience, (of something similar in aircraft), I can support Terry's absolute demand that there should be a low impedance connection between the Earth at the VFD and the Frame of the Motor. Also the cabling should be screened, (and twisted if possible).
I don't have much experience of industrial CT's but aircraft 3-phase CT assemblies are normally mounted inside an alloy screened box, craftily arranged so that the screening extends down the holes of the CT's without forming shorted turns. This avoids capacitive coupling between the feeders and the secondaries of the CT's. The secondary currents are taken off via screened twisted pairs, with those screens down to Earth at the destination only.
In one instance I have seen a screened cable jacketed with a second outer braid. The outer was Earthed at the source and the inner at the destination.
The RS485 isolator should be close to the PC (probably as above already), cabled as 3 (or 5) wire, so that there is a direct 0v-0v connection between the isolator and VFD. This keeps the CMV as low as possible.
Cabling should also be done in screened twisted pair, with the screen connected to the 0v at the VFD.
120 ohm termination at each end, and each receiver should have a local differential R-C low pass filter. RS485 comms can work ok with a surprising amount of low pass slugging. Waveforms can look awful on the scope, but the receivers can still sense the correct logic ok. Run the comms at the lowest speed possible. BTW: Shielded Cat5e comms cable is good for both comms and instrumentation.
It's the other way around for me. I am an RF/analog/EMC guy who sometimes ends up digging through firmware, like when my usual tools found another bus contention and everybody thought it just can't be so. Then I probably feel a similar pain.
Yes, pretty much like Ethernet. It is important to wind them carefully and bifilar. Bifilar means twisting primary and secondary wires about two twists per inch and then winding them onto the core together. For signals in the MHz range I use toroids of #43 ferrite material and for stuff below a MHz mostly #77 material (Fair-Rite, bought via Amidon). The required number of turns increases with circuit impedance and with how low in frequency the signals can be. Use wire with a beefy insulation for better capacitive isolation. Not quite the stuff electricians use, maybe half the insulation thickness and much thinner wire.
Good quality LAN transformers can also work. But they would have to be from a reputable mfg such as Murata.
As Terry said DMMs are not a good tool here. Their leads pick up noise, create loops and the internal circuitry of a DMM might not be designed to handle large electro-magnetic interference loads. If you have the time try the same with an old-fashioned analog meter. The really old kind, no battery and certainly no electronics in there.
In fact I had a DMM die on me after an RF susceptibility test and I was only blasting about 100 watts or so in an RF cell. It was a good name brand in a fancy holster, not some hobbyist version.
I have been in some 'interesting' sw/hw fights myself! I was working tool roads a few years back, and they had an interesting 'hardware' problem. Every weekend, around 2-3 in the morning, the system went crazy. It started a network cascade and every toll system on the road went down. It was 'obviously' a hardware fault, so they called in some network experts to solve the problem. They put a network sniffer on the system, and watched the traffic...
Sure enough, early sunday morning, there was nothing, nothing, nothing, and then a huge cascade! What was the cascade? I bunch of 'Where are you?' messages! They programmers had built in a "If I don't hear from a system for 'x' minutes, check to see if they are still there..." routine. They had NOT built in the stuff to QUIT asking once it got an answer...
If you really are stuck with regular cable there may be another option. It's not as good as shielded and twisted cable but at least a little progress: Run it through metallic conduit. Best would be copper tubing where the pieces are pre-soldered together and grounded.
I don't know how good the flexible metallic EMT is (the stuff that looks like a large shower hose). Probably not very good for RF. If it has to be flexible you can also try thin corrugated copper tubing as is used for connecting a gas appliance to the gas line (hardware store).
Those boards sometimes use isolator chips. While these may isolate well enough to pass safety muster they may not be that good in isolating RF loops. Transformers are the thing for data transfer in a noisy environment.
Flaky comms on site, (in front of the customer), is a pain in the bum, and the most expensive place have to sort the problem. It is generally cheaper to do it properly in the first place, back at base.
Do tell.... Earth Management on big systems is a hard lesson to learn.
When I tested one of my designs for defibrillator safety the PBX system quit and some computers froze up. I got dirty looks despite the fact that I had placed a huge wallpaper on an easel at the entrance asking people to back up a lot that morning.
Or BAT54 duals if you need to cap before the substrate diodes wade too deep into the muck.
Cool. I am surprised they recommend plywood for the applicator side. In California the fire marshall would have a beef with that. Maybe it isn't as hot and dry in Australia as it is here. The transformer could probably be scrapped out of one of those bug zappers. Is "AS" Australian Standard?
No its not, its a high-frequency variable duty cycle "square" wave, whose duty cycle is modulated such that the average output voltage is a
100Hz sinusoid. Who's VFD? (Danfoss is my guess, they like the term VFD)
Gidday,
Drives are nasty. Firstly, ensure you have some form of a co-axial feed from the drive to the machine such as neutral screened cable, with the screen well bonded to the drive *AND* the machine. This is exactly wrong for an instrumentation cable, but right for the drive. Herewith an excerpt from a report I wrote for a customer experiencing this sort of problem in Nov 2003 with a 400kW rock crusher:
"The existing hardware operates in an extremely adverse environment, both physically and electromagnetically. The physical environmental hazards are primarily temperature, vibration and fine particles, which should be assumed to be conductive due to crusher plate abrasion. Electromagnetic hazards are primarily caused by the motor drives used in the rock crusher itself; drives generate significant EMI, and are responsible for the bulk of EMI problems in industry. Ideally drives and motors should be connected using ?screened? cable, typically steel-wire sheathed. The drive cable screen should be firmly bonded to both the drive and the motor ? note that this is exactly the wrong thing to do for an instrumentation cable.
Motor drives generate huge high-frequency (HF) current spikes which flow through the motor winding capacitance to the motor chassis. Both the drive and the motor chassis are earthed, so these HF currents flow through the earth connection and back to the drive. The amount of this HF energy which radiates away is directly proportional to the size of the physical loop the current flows in. Connecting the cable screen as above ensures a good high-frequency connection between the drive and motor ? basically it minimises the physical size of the loop the HF currents flow through, thus greatly reducing the amount of HF energy which radiates away"
Once you have done this (minimised the source of emissions) you can then reduce the susceptibility of your setup. Areas to look at are:
- optically isolate the RS485 link. This is necessary to break the earth loop from PC to drive (which, by Norton, carries some of the HF currents). A moot point if you are running the PC from batteries, but of vital importance when mains powered
- keep wiring loops small. twist, twist, twist. ground plane, ground plane, ground plane. Drives spew out vast quantities of H-field, which your loops will happily receive. Technically this reduces the coupling, rather than the susceptibility, but who cares as long as it works right?
- correctly terminate the RS485 cable
- use high permeability common-mode chokes on the various dangly wire bundles.
As to why your hall effect sensors dont work, I suspect its poor layout/decoupling, or perhaps ratshit DCCTs. The VFD has 2-3 inside, and they work fine. The problems I've come across with current transducers are:
- capacitive coupling from the drive output. the fix is simple, a grounded electrostatic shield around the cable as it passes thru the DCCT. The ground connection must be low inductance. In practice drive manufacturers sample in the middle of each pulse, thereby avoiding the switching edge noise.
- variation of current as a function of cable position, and/or adjacent metalwork. sign of a crappy DCCT....
Make sure your DCCT supply (supplies) are well decoupled right at the DCCT. Again, twist the wires (stranded CAT5 cable is great for this sort of job, esp. STP). Also look at how you measure the DCCT results, and reduce/remove loop area - if with a scope, use a coax connection to the scope/probe rather than a 6" ground clip.
there's a clue in there - DMM. Can you perhaps take a few photos of your setup, and post them to a.b.s.e ( alt.binaries.schematics.electronic ).
It sounds to me like you have some form of CT who's output is monitored by a DMM. The great thing about DMMs is that the leads make fantastic loop antennas, and excel at picking up noise. On several occassions doing drive EMC tests, I have been bitten by attaching DMMs to the D.U.T. - in one instance, we gave up on the DMM to measure smps voltage (which went up 40% when we turned on the 40W transmitter beside it), and just listened to the fan speed instead! Dis-connecting the DMM actually made the job easier.
The problem is to do with the size of the physical loop formed by the leads. This loop is a one-turn inductor, whose inductance is proportional to the loop area. And inductor converts current into H field, and vice-versa (terrible description, but good enough here). When some H field impinges on your loop, the amount of current induced is again proportional to loop area. The DMM is quite high impedance, so tiny induced currents can result in large measurement errors (alternatively, one does not have to supply much current to apply a voltage across the DMM terminals).
Lets just say, for arguments sake, you have a CT with a 10 Ohm burden resistor. This will be connected to the DMM leads - my fluke 12 leads are 1.5m long, enough to make a loop about 1m in diameter, which is sizeable (0.72m^2 or so). what to do?
well, for a start, cable tie the leads together (say every 6" or so). Now the loop area drops to a minimum of 1.5m * 2Ti where Ti is the insulation thickness, say 1.5mm. so the area drops to 0.0045m^2, about
160 times smaller than before. In practice the loop area will be bigger than this, but such an approach can reduce the loop area by 50 - 100x.
That will help, but is it enough? If not, the next step is to twist the wires. If the physical dimensions of the twisted "loop" are very much smaller than the wavelength of the offending H-field (they will be) then adjacent twists have the same H-field across them. However the twisting of wires ensures that the current induced in one loop is in the opposite direction to the current induced in the adjacent loop, so the two cancel. This in effect reduces the inductance even further (and is why pretty much all comms cables are twisted, eg ethernet, phone etc).
This "minimise loops" concept applies everywhere in your wiring (especially PCB layout). Eventually, the loops can be minimised no further - eg the DMM inputs are a fixed distance apart, and away from the DMM internal circuitry - aint much you can do about that. So at some point the unavoidable loops dominate noise pickup, and no further gain will be had from twisting wires.
once the loops are optimised:
If the CT has a 10 Ohm burden resistor, then paralleling another (say)
1k Ohm across it will have a negligible effect on the measurement. So slapping a 1k directly across the terminals of the DMM wont bugger up your measurement, but will make it a lot harder to force a noise voltage across the DMM terminals.
In practice, I would use a BNC-to-banana converter plugged into the DMM, with a 50 Ohm coax thru-terminator plugged into that (I have a whole bunch. I normally get people to use a T and an end-terminator, as its about 10x cheaper than a thru-terminator, and unless you are at very HF, its good enough), and a 50 Ohm coax cable to the CT, which has a BNC socket soldered to the burden resistor, and bugger all loop between the R and the CT (and the R + BNC). Have a look at a pearson CT...
(and yes, I'd have to re-do my CT output scaling calcs, cos 50 aint >>
10, but the maths is easy)
why dont CM toroids help? the noise picked up by your loop is differential mode not common mode.
why dont LPFs help? the DMM is picking up noise on its leads, which effectively gets added to the LPF output.
If it makes you feel any better, everyone else has these problems with VFDs. An amusing anectode:
A customer of ours built a large computer-controlled sawmill - really cool, a big bank of paralleled PCs solved a travelling salesman-like problem to optimise the log cuts to maximise return in real-time based on market data. Fancy sensors, lasers etc. They had a fancy position sensor that is basically a plunger with a magnet on the end, in a pipe fed GHz RF at the other end. Some miraculous electronics & a dead white guy meant they could have an absolute accuracy of 0.1mm or so over tens of meters. But it used H field to work.
They had a lot of big drives (>= 250kW), and every time the drives turned on the whole system went apeshit - rams firing off, 12' radial arm sawblades whizzing out and back, that kind of thing. Because all the position sensors started firing off nonsense, and their controller wasnt very robust (no built-in bullshit detector). The cables were screened, but the screen was *NOT* connected to the motor, only the drive.
So they got hold of our service dept, who told them to tie the cable shields to the motor as well. Their engineer was a young (my age) instrumentation guy, and thought that was wrong, so for 3 months he didnt do it. Eventually, after repeated calls to the service dept. he did, and *voila* all the problems went away.
This guys manager wanted to know what was wrong with our drives and why this fix worked (and why his guy didnt implement it for so long), so I got sent there (with our service manager), to explain why it worked and that *all* drives behave the same. In the nicest possible way, I had to explain to the dozen or so customers why this guy was wrong - actually I just explained the physics behind the interaction, and the physics behind this guys erroneous decision, which was right for instrumentation, but wrong for drives. And man, did his boss rip him a new one afterwards, it was embarassing to see.
No harm was really done though, because during this 3-month period they were commissioning the entire plant. It was only when all else worked and the drives had become a bottleneck that the heat went on this guy, who finally did what we asked.
Thats the safe way to do it. The twisting on the CT cable takes care of H-field, and the single-ended shield deals with E-field. I can see why they wouldnt muck about in aircraft :)
that doesnt strike me as a good idea for a drive output cable, but I have seen it in instrumentation. Betcha its a pain to terminate :). Tsaliovich's book has some pretty pictures in it re. cable shielding effectiveness.
Yep, I'd agree with all of that. To be honest I have never had a problem with RS485 and drives, but to be fair I started work at a drive manufacturer, and they already knew all this stuff, which I picked up very quickly - its nice to not have to make all the mistakes yourself.
So I went and found a whole different bunch of mistakes to make....
On large systems and even on Phillip's drive system one often has to accept the rule that there simply is no common potential between individual modules. From a 50/60Hz point of view, hopefully yes, but for anything much above that all bets are off.
hear hear. I've had some hilarious hardware/software arguments too - me and the s/w guy basically saying "its your fault". Usually its s/w, perhaps 75% of the time - I suspect because its so easy to be careless with s/w, and so hard to spot (not looking is also the most common technique used for s/w peer reviews and testing).
That we can usually do calculations to prove the efficacy of our designs is a huge bonus to us, along with spice, design-rule checking etc. A whole host of reasons not to write firm/software.....
bugger. I havent killed a DMM yet, but I did have one of the leads on my fluke 87 break while I was testing a 50kW regenerative rectifier. Suddnely the sytem went bonkers, it took about half an hour to figure out my DMM was lying and that it all worked fine. Funny how we always assume the worst with complex things. Customers are like that too - if something is wrong it must the the drive, but is usually the wires :)
a few years back I did my limited electrical registration (allows me to fiddle with 3-ph fixed wired 400Vac doodads). The instructors were at pains to recommend test-measure-test with DMMs for exactly that reason, a strategy I immediately adopted. I fitted a relay to a VFD in a milking shed yesterday for the local sparky, who laughed at me when I did the TMT. But I'm still here.....
we had a showering arc generator for EMC susceptibility, and every time we turned it on the R&D/production door swipe-card died, so all foot traffic had to go thru the power lab. The mfg eventually fixed it. The first time the SAG was powered up, the photocopier died too, so we bunged a line filter in front of it and that kept it alive. Phone conversations were tricky with the SAG on too :)
The SAG is a great little toy, easy to make, fun to use and highly destructive. Rather than using a coupling clamp, we attached a multimeter probe to the output, and probed directly to the I/O terminals. We could write our names on the alodised metalwork with the spark. One product I designed had no clamping on one input, just an 0603 cap. the SAG made the cap flash over, emitting bright flashes - caught a glimpse, turned the lights out and voila, weakness glows in the dark. BAV99 to the rescue....
If you're interested, I posted the PDF showing how to make one to abse.
ayup. You've done this before.... I once managed to design a circuit that occassionally triggered the parasitic supply SCR in a lattice CPLD. The numbers would burn off the chip, and a little dome would swell up in the centre of the PLCC. oops.
I used a neon xfmr fed from a step-down xfmr to get the right voltage (it was free). When I say "I" I mean my slave ^H^H^H^H^H^H technician. He did a much nicer job than I would. I think we made the frame from formica, but your right, plywood might not be such a good idea. Mine is in a big Al case, so it can catch fire inside if it wants to :)
Yes, AS/NZS. We often pinch the standard numbers from europe/usa....
oh yeah, and I've sure seen some terrible comms links. 50m of RS232 cable, both devices earthed with a bitwise protocol sans parity - one bit = one task = noise sample & hold & act erroneously :)
Now thats something I'm interested in hearing more about...
Beautiful. And fits nicely with my theory that many programmers never test anything, and seldom if ever think about failure modes. I bet you've seen numerous systems hang when no response is received...
the funniest one I have had was a drive we designed using an existing micro & software, with all new hardware. We re-arranged all the ports, and vandalised the software to suit. All worked spiffingly well, until about 3 days after the product launch, when a customer complained to the service dept his drive kept tripping when he used the 4-20mA input. After the inevitable argument with Mark, I did some tests and found that at about 10mA the drive tripped. I configured the input as +/-10V, same problem at about 0V. It happened when the uC ADC pin reached about 2V or so, coincidentally the logic threshold. So I looked at the original product, where that pin was the emergency stop switch input. Armed with this ammo I re-started and easily won the argument. Mark went and had a nosey, and came back a few minutes later looking sheepish, with the problem solved. He'd re-written the I/O code, but had forgotten to remove the old E-stop code (all the rest was gone). The 80C196 was set to use that pin as an ADC, but the digital input SFR still worked :)
(best not to think about the 3 goes I had my tech have at getting a LED to light up a few weeks later)
the worst one was when a programmer set up a command to shut down the comms link, and save that state to EEPROM so it wouldnt turn back on again. The unit is gooped and screwed into an IP68 box with only the comms link & power coming out, and to get it to talk again the lid had to be removed (12 screws) so we could access the diagnostic serial port. That wouldnt be so bad, but another programmer was upgrading a customers screen (1000 units) and sent out a (untested) broadcast "shut up forever" command. Luckily the broadcast didnt work so well, and he only "killed" a few hundred modules. But a guy in a climbing harness had to undo a thousand screws to remove the crippled modules, then a few thousand more to revive them on the ground, and of course put it all back together again.
We made ourselves look like complete idiots to the customer so much for the "image quality upgrade". We had the screen back and running that day, which kept him happy, but if it had been a game day we would have been in the shit, it took hours to fix. When asked why he saved the state to EEPROM, the programmers reply was he had a choice (to save or not to save) and "just" chose to save it. The server was updated that day (*about 15 minutes after we tracked down the root cause) to never allow a broadcast of that command - until then it was transparent, you type it, it does it. The firmware was also hurriedly modified to remove the built-in self-distruct command :). And the programmers got an abrupt lesson in "thou shalt not f*ck with the customers equipment"
I have read here before about this business of putting different signals into the drive. Why is this? I think last time someone mentioned (sphero perhaps) injecting a 3rd harmonic or something of the like.
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.