DIY Electric Car Forums banner

Anonymous Q: Was it wrong to start this thread?

  • Yes, it unfairly tarnished Kelly.

    Votes: 4 12.1%
  • No, it overall provided useful information.

    Votes: 29 87.9%
1 - 20 of 174 Posts

·
Registered
Joined
·
3,218 Posts
Discussion Starter #1
Greetings all. First post here. I've been contracted to design a dc motor controller by an electric vehicle conversion kit manufacturer that currently uses, albeit reluctantly, Kelly brand controllers. Through trial and error they have found that 96V is the maximum safe battery voltage that can be used with the KDH12600 before they start going up in smoke. I took one of the blown controllers with me for a post-mortem analysis and am sharing what I've found here just for the helluvit. I will post successive pictures in individual posts to keep each post reasonable in size.

First picture is a close-up of the MOSFETs used in this controller, Fairchild part # FDP2532. The black stuff is polyurethane conformal coating. It can be removed in spots with a hot-air gun such as used for SMT rework and with various solvent-based formulations. There are 24 total of these MOSFETs arranged in a half-bridge formation of 2 banks of 12. The salient specs for this part are 150V; 56A (@100C); 0.016R; 107nC Qg[max]. As 12 are used in parallel this controller can theoretically meet its 600A specification though with the caveat that few TO-220 parts use bond wires capable of handling 50A on a continuous basis. At 50A each FDP2532 will dissipate 40W which is a reasonable amount to ask of a TO-220 part that has been attached to a nominal size heatsink (say 1.5C/W).

Unfortunately, there is no transformer or Hall-Effect device for sensing total current but since there is over-current protection I would surmise that it is by measuring the voltage drop across Rds[on]. This would go a long way towards explaining the reports that in otherwise identical systems a Kelly controller can't deliver nearly as much current as others'.

Doubly-unfortunately, the 150V Vds rating of the FDP2532's pretty much dooms them to failure if the motor leads are more than a couple feet long and/or the battery pack voltage is more than 96V. And if that wasn't tempting fate enough, there does not appear to be any sort of snubber to suppress spikes from stray inductance.

Continued...
 

Attachments

·
Registered
Joined
·
3,218 Posts
Discussion Starter #2 (Edited)
...continued.

The capacitor bank is composed of (10) 330uF/200V capacitors, series KXG by United-Chemicon. This a relatively good choice of capacitor, but there could probably stand to be a few more of them, though. This is because they can each sustain 1.43A of ripple at 10kHz and 105C so a bank of 10 is good for 14.3A of ripple. There is no precise value of capacitance necessary because it depends on the motor current AND the motor inductance. Generally, though, higher current motors have lower inductance and both these things demand a higher ripple current rating of the capacitor bank. To figure out how much inductance is needed to prevent exceeding the ripple current rating of the capacitor bank you rearrange the inductance equation, V=L*(dI/dt) to L=(V*dt)/dI. Worst case ripple occurs at 50% duty cycle so if the controller regulates average motor current the ripple at 50% duty cycle will be 50% of the maximum allowed motor current (e.g. - 300A,perhaps, for this controller, though that setting can be changed in software). Thus, plugging in the numbers we get L=(108V*32uS)/300A = 11.52uH. That seems to be a reasonable minimum inductance requirement to me.

Continued... (gotta go take care of a few things - will post more later).
 

Attachments

·
Registered
Joined
·
3,218 Posts
Discussion Starter #4 (Edited)
...continued.

The label on the KDH12600 indicates it can be used on 24-120V and deliver up to 600A. The 600A rating can be met for the semi-customary 1 minute given the specs of the MOSFETs and the heat sink but I am highly skeptical this controller would survive for long at 108V, much less at 120V. It basically comes down to stray inductance (10-30nH/inch) and the switching speed of the MOSFETs. Even at a rather leisurely transition time of 500nS (0.5uS), 600A will induce somewhere between 12 and 36V per inch of wire in voltage spike. At a minimum, one should employ an RC or RCD snubber on each leg of the half bridge AND use switches rated for 2x the maximum supply voltage.

Also note that the est. worst case dissipation from conduction losses, only, at maximum current and duty cycle will be 480W: (600A^2) * (0.016ohms/12).

I'm pretty sure that the microcontroller used in the KDH12600 is a Freescale MC9S08AW16MFUE (see Kelly-5.jpg). I'm not 100% sure because an attempt was made to obscure the part number by taking a Dremel to it - especially hilarious given the pedigree of the controller. Someone more skilled at hacking/programming but who was stumped by the conformal coating might be able to use this to their advantage....

As seen in Kelly-4.jpg, the wires from the two circular connectors are not shielded in any way from the intense magnetic ("H") field produced by the controller when switching even modest amounts of current (say, 50+ amps). Such EMI *could* easily cause the microcontroller to hang, but there are ways to prevent that in-circuit.

Finally, Kelly-3.jpg shows the underside of the controller. The most interesting thing to note here is that while all of the drains are bolted to a copper buss bar, all of the sources are merely soldered to the PCB. If this PCB were clad with 2oz copper (and this one definitely uses 1oz), the trace for the sources would need to be 13.6cm wide (~5.4") to carry 600A even allowing for a 30C temp rise!?!

Finally, there does not appear to be any isolation between the microcontroller and the MOSFETs, hence why a separate, isolated power supply is needed. So, when one of the MOSFETs failed (see the first post), likely from just the sort of voltage spike described above, the full force of the battery pack took out the microcontroller. The sheer breadth of destruction something like this inevitably causes, along with the board being coated with black polyurethane means the Kelly KDH12600 is, sadly, not practically serviceable.

Finally, my rough estimate of the parts costs, not including the enclosure (as I am not as familiar with the costs of such) is $200US. This is about right given the rule of thumb for manufacturing that says you should aim for a retail price of between 3x and 5x the parts costs (to account for labor costs, administrative, profit, etc...) *Added: the price of electronic components when purchased in large enough quantities is highly negotiable - my off-the-cuff estimate is just that, off-the-cuff.)
 

Attachments

·
Registered
Joined
·
73 Posts
WOW!! excellent narrative! I think U have now revealed to all the quality of Kelly controllers. There have been several threads indicating interest in purchasing this brand of controller because it was "more reasonably priced". Unfortunately, or fortunately, the old saying holds true: U get what U pay for.
I might add: ALA U know what U are paying for. U just can't beat a good track record and good customer service :cool:
 

·
Registered
Joined
·
293 Posts
I read with interest your comments (& others, too) about controllers that “die the death”. Yes, my Curtis 1231 / ADC 9” combo also had a sudden demise.

If I may add some observations from myself & others, I sense that a major part of the problem arises from the motor impedance (which you noted) & length of motor cable. Thus motor & controller need to be physically close together to keep cable as short as possible. Also, different brands & sizes of motors must be specked for COMPATIBILITY & this is not to be overlooked whatsoever (IMHO). An appropriate diode must be placed across the MOSFET/IGBT to safely stop any voltage that develops from arcing or motor impedance. Hence the need to oversize the volt & amp ratings of the various components to accommodate a multitude of possible real-world scenarios. Hence the big OEMs learned(?) early on from their warranty returns & made numerous revisions while the new guys on the block have yet to graduate from the “School of Hard Knocks”.

To illustrate the damage arcing does upstream of the actual short, here is an experience of mine:
Our company had just purchased 10 vacuum coating machines with state-of-the-art 0-10KV power supplies. We ran them at 6-7KV, yet one by one they were failing. The OEM repaired them at no charge for the first year, after that I replaced a failed $10, 15KV “door knob” capacitor. The OEM determined that the chemical we were vaporizing was causing abnormal arcing within the chamber, so reduce the voltage AND shorten the 17’ length of the Hi Voltage wire. The wire was shortened but voltage was not altered as it would affect cycle times. That mod helped but did not eliminate repeat failures.
Also we had at least 10 electric forklifts in use at the plant, but not once in 11 yrs did I hear of controller failure. Is it due to use of 48V or less? As for EV’s with 120+ volts, I believe it’s a different story & I am led to believe it is due to this induced voltage spike that prematurely kills IGBTs/MOSFETs if not sufficiently quenched. Even when relay coils are turned on & off, engineers know to add a clamping device to limit the resultant inductive spike (coils release voltage when de-energized. Ex: automotive spark-plug coil).

May we continue to learn from each other…

94 S10 120V
 
G

·
Guest
Joined
·
0 Posts
Have you relayed any information to Kelly Controller of what you have found and what can be done to make a better product? If you have not then I suggest you do so they can make the changes. It is in their best interest to have a viable product and a wide open market. If you don't then I guess you will help kelly die an early death. It could be a great controller if helped. Not all of us can help and my controller is still alive. Mine is not a high power controller but it is 72 volts and 600 amps. It is also a SepEx controller. So far so good. It got a tad hot but still works. I need a good heat sink.

Pete : )
 

·
Registered
Joined
·
3,218 Posts
Discussion Starter #9 (Edited)
First off, thanks for the comments. Secondly, my intent here was not so much to bash Kelly (I tried to point out the good and the bad) but rather to use their controller to illustrate the sorts of things one needs to pay attention to if you want to "roll your own".

gottdi: I suppose it bears repeating that I am being paid to design a competing product to Kelly's, so I have no economic interest in helping Kelly out here... ;) That said my opinion on this in general is that, Kelly, like any other company trying to make a name for itself, ought to be pro-actively solving problems with their products, not relying on paying customers to do their beta-testing. Just my $0.02.

david85: thanks for the welcome. While I also have an economic interest in not sharing too much of what I'm working on, I will share some of it.
 

·
Registered
Joined
·
3,218 Posts
Discussion Starter #10
...
If I may add some observations from myself & others, I sense that a major part of the problem arises from the motor impedance (which you noted) & length of motor cable. Thus motor & controller need to be physically close together to keep cable as short as possible. Also, different brands & sizes of motors must be specked for COMPATIBILITY & this is not to be overlooked whatsoever (IMHO)....
There are a lot of trade-offs involved here but while it is overall best to keep the motor and controller as close to each other as possible, the two can be safely separated as long as there is tight magnetic coupling between the current-carrying wires! That is to say, keeping the wires parallel (and preferably twisted together) from controller to motor will allow you to safely extend the distance between the two devices.

Furthermore, a higher inductance motor/wiring system will result in higher voltage spikes during switch (ie- MOSFET or IGBT) turn-off but it also will slow the rate of rise of current if, for example, the motor was locked-up or otherwise severely overloaded. This gives the controller more time to shut down drive to the switches before the current flowing through them results in failure. BTW - you typically have 10uS to turn off an IGBT in the event of a short-circuit before it is destroyed. Most of the microcontrollers used in motor controllers take 20uS or more just to execute the instructions pursuant to a fault interrupt which is why protecting against short-circuits needs to be done "in hardware" right at the switches (e.g. - desat detection). Let the microcontroller monitor and control the *average" current flowing through the switches (i.e. - throttle response).

At any rate, all of this actually reinforces your argument that motor and controller need to be matched. Unfortunately, few motor manufacturers bother to specify the inductance of their motors while few (any?) controller manufacturers specify what range of inductance their controllers can accommodate. An analagous situation would be if a gasoline engine didn't tell you what octane gasoline it needed and gas stations didn't tell you what octane ratings they sold! Crazy, isn't it?

There are many other considerations but they probably belong in a separate thread.
 

·
Registered
Joined
·
1,760 Posts
First off, thanks for the comments. Secondly, my intent here was not so much to bash Kelly (I tried to point out the good and the bad) but rather to use their controller to illustrate the sorts of things one needs to pay attention to if you want to "roll your own".
Indeed you have pointed out several important things to consider and since I've at least been toying with the idea (Kelly seems too cheap, Zilla is out of business so it's either trying to make one myself or get a squealing Curtis), at least for EV number 2 (if number 1 ever rolls out), it's good to not run into the obvious pit falls. That Kelly doesn't even have a snubber diode in their controllers is just bizarre, I thought that was common knowledge that anything silicon based driving anything with an inductance MUST handle the kick back one way or another.

There are a lot of trade-offs involved here but while it is overall best to keep the motor and controller as close to each other as possible, the two can be safely separated as long as there is tight magnetic coupling between the current-carrying wires! That is to say, keeping the wires parallel (and preferably twisted together) from controller to motor will allow you to safely extend the distance between the two devices.
You're full of wisdom! That's a great advice, it sounds so simple when you point it out, yet I haven't even reflected over it. I've planned to run the wires in parallel from the pack to the controller (to try to avoid the car stereo picking up the noise), yet I've not even considered the motor wiring. But then, I haven't even started building... :(

Most of the microcontrollers used in motor controllers take 20uS or more just to execute the instructions pursuant to a fault interrupt which is why protecting against short-circuits needs to be done "in hardware" right at the switches (e.g. - desat detection). Let the microcontroller monitor and control the *average" current flowing through the switches (i.e. - throttle response).
Using a comparator triggering an interrupt in an AVR micro controller running at maximum speed (16 or 20 MHz depending on version) should be able to react within 1 us. Add to that slew rate etc, but it should be possible to handle it in software well within the 10 us-limit as long as you don't try to use the AD for measurement. Not in any way saying that you're wrong that the controllers usually don't do it fast enough, just pondering the fact and considering the options. It SHOULD be doable! I'll let you know as soon as the smoke clears again... :D

Great postings! It sure points out several pit falls no matter if you try to build your own controller or are in the process to mate controller or motor.
 

·
Premium Member
Joined
·
882 Posts
Do you think the same inductance issue is a problem at lower voltages (i.e. will this issue scale down or is it a purely post 96V phenomenon). I have a 72V 400A controller which I am aiming to use at 76.8V and ~300 amps. Other than keeping the cables short what can I do to minimise the risk of such failures?
 

·
Registered
Joined
·
1,760 Posts
Do you think the same inductance issue is a problem at lower voltages (i.e. will this issue scale down or is it a purely post 96V phenomenon). I have a 72V 400A controller which I am aiming to use at 76.8V and ~300 amps. Other than keeping the cables short what can I do to minimise the risk of such failures?
I bet Tesseract can give loads of good recommendations on this, but one thing I definitely would do is to check if the controller does snub the kick back or not. If it lacks the diode, add one externally! It will definitely make the system more robust and decrease the risk that your controller burns up.

Before this thread I'd expect all controllers to do that (as I've said before I thought that was common sense), but apparently that's not always the case. If the kick back isn't loaded down, theoretically it will reach infinite voltage. In reality there will of course be a limit, something's gonna ground the spike sooner or later, the question is only what, at what voltage and if it'll damage something in the process...
 

·
Registered
Joined
·
299 Posts
Thank you for such good info. What specs would you look for in a good snubber diode on a 120v 400a controller. I will soon be testing my homemade 400a mosfet controller on my motorcycle at 72v and had completely forgotten a snubber. Thanks again for your recommendations.
 

·
Registered
Joined
·
1,760 Posts
Thank you for such good info. What specs would you look for in a good snubber diode on a 120v 400a controller. I will soon be testing my homemade 400a mosfet controller on my motorcycle at 72v and had completely forgotten a snubber. Thanks again for your recommendations.
Oups! That could've gone sour. :(

Well, reverse voltage obviously will have to be more than the intended drive voltage, but on the other hand diodes that can handle 200-300 Volt isn't really expensive so I'd go for overkill. a quick search on Farnell gave me for example a Fairchild FFH30US30DN, a dual diode that can handle 2*30 (70 repetetive peak) Ampere and a maximum of 300 Volt reverse voltage.

I'd say that 5-10 of those (they only cost a few bucks each so why not go for overkill?) would probably handle it without problems, but I'm in no way an expert at switching electronics so please verify this before going all kamikaze on my words only... :D
 

·
Registered
Joined
·
3,218 Posts
Discussion Starter #16 (Edited)
... That Kelly doesn't even have a snubber diode in their controllers is just bizarre, I thought that was common knowledge that anything silicon based driving anything with an inductance MUST handle the kick back one way or another.
I'm not sure 100% what you are referring to, but I suspect you are actually referring to the freewheeling diode that goes across the motor in 1-quadrant (no regen) controllers... No such controller would survive one switching cycle if this diode wasn't present - it conducts the motor current during the switch off time, allowing it to decay smoothly. A "snubbing" diode is used, along with a resistor and a capacitor, across the switch to form what is called an RCD snubber (also referred to as a dV/dt snubber because it slows the rate of voltage rise). This is the sort of snubber I was referring to. A mismatch in transition times between the switch and the diode, along with the stray inductance of the wiring in between them, is what leads to voltage spikes that kill one or both. For example: a rule of thumb is to assign 20nH per inch of wire (this can vary quite a bit, obviously). There's about 7" of total distance between one end of a MOSFET bank inside the KDH12600 and the other, so approximately 140nH (0.14uH) which seems neglible, really, but if you switch 600A across that 140nH in 200nS you'll produce a voltage spike of 420V. The spike has to go somewhere, usually by avalanching the opposing bank MOSFETs (after all, they will be forced into conduction if more than 150V is applied to them.) A lot of MOSFETs are designed to handle this - the FDP2532 is avalanche rated, in fact - but it is still harmful to the device and considered really sloppy by most engineers.

In switching power supplies the snubber design is arguably the single most critical aspect; throwing around even more current, as in a motor controller, only makes it more so.


Using a comparator triggering an interrupt in an AVR micro controller running at maximum speed (16 or 20 MHz depending on version) should be able to react within 1 us. Add to that slew rate etc, but it should be possible to handle it in software well within the 10 us-limit as long as you don't try to use the AD for measurement. Not in any way saying that you're wrong that the controllers usually don't do it fast enough, just pondering the fact and considering the options. It SHOULD be doable! I'll let you know as soon as the smoke clears again... :D
Yeah, I imagine there are a bunch of ways to optimize the code which may make this do-able. Programming those little buggers is definitely not my strong suit - I have a stack of books on them but haven't managed to get through a single one... FWIW, I am looking at the ATTiny461 for my controller design as it has the fast hardware PWM built in. Not sure if I will be able to cram all of the code needed for a 2-quadrant motor controller (forward motoring and regeneration) into the 4k of memory... maybe if I leave out all of the comments... ;)
 

·
Registered
Joined
·
299 Posts
Forgive my ignorance of terms but is a snubber the same as the freewheel diode? I have twelve MBRF40250TG Schottky freewheel diodes that are rated at 40a 250v that are wired in parallel with the motor. They are mounted to the heat sink.
 

·
Registered
Joined
·
3,218 Posts
Discussion Starter #18 (Edited)
Forgive my ignorance of terms but is a snubber the same as the freewheel diode? I have twelve MBRF40250TG Schottky freewheel diodes that are rated at 40a 250v that are wired in parallel with the motor. They are mounted to the heat sink.

No. See above post and refer to the attached image. Components R, C and D comprise the snubber. If the freewheel diode is replaced with a switch then you have a 2 quadrant drive (i.e. - motoring and regeneration in one direction only).
 

Attachments

·
Registered
Joined
·
1,580 Posts
Out of curiosity, would a Kelly REGEN controller be more robust since it deals with the back emf and reverse motor input?

I have a Kelly KDH09401 with regen on my series wound motor. The REGEN is disabled since I am not using a PM motor, but so far I have not have a problem. I have only run it on 60 volts tho because I don't have the front rack built for the rest of my 96 volt system.

Maybe it helps that my motor is an egg-beater (6.7 inch ADC K99-4007) and probably has a small amount of inductance compared to a 9 inch motor. Man do I need and oscilloscope......
 

·
Registered
Joined
·
3,218 Posts
Discussion Starter #20 (Edited)
TheSGC - no, the amount of motor inductance is only relevant to how much reservoir capacitance you need (lower inductance = more capacitance) and how fast your current-limiting circuit needs to respond (lower inductance = faster response needed).

It is the inductance in series with the collector-emitter or drain-source connections that is the problem. Literally, the stray inductance of the circuit board or interconnect wires/buss bars inside the controller. The reservoir capacitance serves to buffer the controller from the stray inductance of the Battery (+) and (-) cables; an RC or RCD snubber is needed across the switches collector-emitter (drain-source) terminals to deal with the strays after the reservoir capacitors.

Once again, the length of the cables from controller to motor is not terribly critical as the inductance of the motor will undoubtedly greatly exceed the inductance of the cables (especially if the cables are in parallel and twisted together - understandably a tough task for double-ought (00) cables). The main reason you want to keep them as short as possible is because they can be incredible H-field (magnetic) antennas spewing EMC all over the place causing other devices - like the microprocessor inside the controller itself - to go haywire.

Oh, and yeah, unless you are extremely lucky in life, you'll definitely need an oscilloscope to build a controller with half a chance of working reliably. I'm collecting my thoughts on such right now for posting in a separate thread that I'll title, "So you want to build a motor controller, huh?" In it I'll address some of the things to consider before blowing up thousands of dollars worth of parts, getting one or both eyes poked out, etc...
 
1 - 20 of 174 Posts
Top