Doc EXTREME 36 FETS controller

Pat,

It's always nice to have someone else look over the math. I'm not sure if this was my primary source during that previous discussion, but Fairchild Semi has a nice app note on making these calculations:
http://www.fairchildsemi.com/an/AN/AN-6005.pdf

Since we're discussing cooling, we need to make sure we break down exactly how the losses are distributed across the various FETs. I think that you can assume that commutation is happening much more quickly than temps are changing, so the high-side losses are averaged across the 3 banks and the low-side are distributed across the 3 banks. I expect that means the low-side will end up hotter, but the calculations will tell. The limit of the system will be determined by the hottest bank, so it's not accurate to take the total loss and assume it's evenly spread across all FETs.

If you want to be really particular, you could see what happens in the worst-case which would be operating at zero speed - so no commutation changes. In that case one of the high-side banks would see all the high-side loss (3x the average power). On the low side, one bank would see all the diode loss and a different bank would see the conduction loss, so worst-case would be the higher of those two (almost certainly the diode loss). If you can make the cooling effective enough to handle these loads, then you should be fairly well-protected against popping a FET. At least, due to thermal stress. :-D
 
That thread is becoming really interesting about thermal desing guys!

I am now searching for those really good BerGquist sil pad ( the pink one)

Apparently they are the best.

I guess it's the Sil-Pad 1500ST, (8 mil thick)

.. i'm still searching for the best i can .. but there is alot of parameters to examine!!
http://www.bergquistcompany.com/thermal_materials/sil-pad.htm

I still have the grey sil-pad sheet that i got with few infineon controllers, but i dont know what is their thermal conductivity.. I guess it's over 3degC/W...

Doc
 
Doc,
I saw in the later posts something I missed earlier...that your copper heat spreader is only 20mils thick. This will have almost no effect as a heat spreader (just too thin for any real horizontal heat flow) and in fact may raise the thermal resistance even higher since it's thin/flexible enough to not make 100% metal-to-metal contact with the Al bar.

I highly recommend either not using the copper at all (which solves your soldering problem) or going to a much thicker copper bar...at least 100mils-125mils thick. If doing this, as mentioned earlier, the soldering is pretty easy if you use solder paste and reflow the FETs onto the copper using a toaster oven or even over another piece of metal atop your stove. Then it doesn't matter how thick the copper is. Paying attention is key if not using a temperature-controlled hot plate (or similar), but it's definitely doable.

But, since you don't have the room to use much copper, I'd recommend not using it at all. While effective heat spreading can be helpful, its impact on the overall thermal resistance of your design probably isn't worth all the grief of soldering the FETs to a spreader and then trying to make sure it's 100% flat against the aluminum bar. IMHO, it's definitely not worth doing for a 20mil thick spreader.

The SilPad 1500ST stuff is pretty good but have you considered just using 0.5mil mylar (with thermal paste) instead? It's tough, incredibly thin, and outperforms even the best pads due to it being so thin. And its dielectric strength is plenty high, even at 0.5mil thickness. Just ensure that your aluminum bar is flat and smooth (having it milled on the one face is a great way to do this) and you're set with an effective, easy to use heat sink. :)

The paste can easily be applied to the aluminum bar in an almost see-through-thin layer (less than 1mil thick!) and the mylar sheet applied to it. The paste will hold the mylar sheet in place. Then apply paste to the back of each FET and mount them normally.

If the aluminum bar can't be milled, place a sheet of 1200 grit paper on a mirror (it's perfectly flat), grit-side up, and place the bar FET-mounting-side down onto the grit paper and move the bar around in a circular pattern for a bit. Then take a look at the mounting side and see where the grit rubbed. If the entire face of the mounting side of the bar has been touched by the grit, that's perfect, you're ready to go! If you see lots of spots not touched by the grit, and they're where FETs might be located, you might need to go to a coarser grit to flatten the surface and then to progressively finer grits to smooth the surface.

After doing that for a bit though, hauling the piece of metal to a machine shop for a quick milling/lapping starts to sound real good. :mrgreen:
 
CamLight said:
Doc,
I saw in the later posts something I missed earlier...that your copper heat spreader is only 20mils thick. This will have almost no effect as a heat spreader (just too thin for any real horizontal heat flow) and in fact may raise the thermal resistance even higher since it's thin/flexible enough to not make 100% metal-to-metal contact with the Al bar.

I highly recommend either not using the copper at all (which solves your soldering problem) or going to a much thicker copper bar...at least 100mils-125mils thick. If doing this, as mentioned earlier, the soldering is pretty easy if you use solder paste and reflow the FETs onto the copper using a toaster oven or even over another piece of metal atop your stove. Then it doesn't matter how thick the copper is. Paying attention is key if not using a temperature-controlled hot plate (or similar), but it's definitely doable.

But, since you don't have the room to use much copper, I'd recommend not using it at all. While effective heat spreading can be helpful, its impact on the overall thermal resistance of your design probably isn't worth all the grief of soldering the FETs to a spreader and then trying to make sure it's 100% flat against the aluminum bar. IMHO, it's definitely not worth doing for a 20mil thick spreader.

The SilPad 1500ST stuff is pretty good but have you considered just using 0.5mil mylar (with thermal paste) instead? It's tough, incredibly thin, and outperforms even the best pads due to it being so thin. And its dielectric strength is plenty high, even at 0.5mil thickness. Just ensure that your aluminum bar is flat and smooth (having it milled on the one face is a great way to do this) and you're set with an effective, easy to use heat sink. :)

The paste can easily be applied to the aluminum bar in an almost see-through-thin layer (less than 1mil thick!) and the mylar sheet applied to it. The paste will hold the mylar sheet in place. Then apply paste to the back of each FET and mount them normally.

If the aluminum bar can't be milled, place a sheet of 1200 grit paper on a mirror (it's perfectly flat), grit-side up, and place the bar FET-mounting-side down onto the grit paper and move the bar around in a circular pattern for a bit. Then take a look at the mounting side and see where the grit rubbed. If the entire face of the mounting side of the bar has been touched by the grit, that's perfect, you're ready to go! If you see lots of spots not touched by the grit, and they're where FETs might be located, you might need to go to a coarser grit to flatten the surface and then to progressively finer grits to smooth the surface.

After doing that for a bit though, hauling the piece of metal to a machine shop for a quick milling/lapping starts to sound real good. :mrgreen:

Thanks John,

I thought that the copper sheet would improve the current sharing and also help the center pin of all fets to get better current share so each of the 6 fet in each bank could heat more equally... no? About the thermal conductibility, i would have used the sil-pad between the copper and the aluminum... so it could fill with good uniformity?

The 0.5 mil mylar seem a great idea, but the less thick the material is(0.5 mil), the more it can break with the sharp edges of the mosfet? I mean i know they are often sharp .. what is your opinion about that?

I have acces for milling job at work.. and i consider that it's a must since i soldered the copper tube on it and it is now not perfectly flat on the back side.

Also, I already lapped aluminum with special lapping tool.. and also ever tried with mirror.. but at work we have a flat marble and some granite ultra flat optical table that are not used... so i have some good references flat ( we used them with some Zygo optical interferometer :wink: . If only i could have the patience to make it flat like a mirror.. but the This is a 'L' bar and not a flat bar.. so the polishing movement can just be made with little circular movements...

Anyway.. i'll do my best.

Oh.. and why Mylar instead of Kapton?
 
rhitee05 said:
Pat,
It's always nice to have someone else look over the math. I'm not sure if this was my primary source during that previous discussion, but Fairchild Semi has a nice app note on making these calculations:
http://www.fairchildsemi.com/an/AN/AN-6005.pdf
I'm really doing it for my own education since it does look like those equations do give valid results (exept at 0% duty... ;), but that just almost doesn't make sense anyways right?). I am going to separate things out for clarity, for the input variables, the intermediate results and also add some extra output details (like splitting the low side losses into two parts). I also just installed the Open Office suite, so it'll let me play around with the spreadsheet app at the same time. I would also like to add the thermal variables such as we've been discussing here to help get a quick idea of rough junction temps. I even have an excel spreadsheet already made up that uses MOSFET current to calculate the final die temp depending on thermal variables (this is an iterative relationship because of the FET's die Temp<-->RdsON feedback process for any given current). Using this thing made me very conscious of how a low Rth case to sink junction is so important at high current levels!!


Since we're discussing cooling, we need to make sure we break down exactly how the losses are distributed across the various FETs. I think that you can assume that commutation is happening much more quickly than temps are changing, so the high-side losses are averaged across the 3 banks and the low-side are distributed across the 3 banks.
I'm not so sure about this being true even at highish speeds based on the IRFB4110's datasheet, page 5 [Maximum Effective Transient Thermal Impedance, Junction-to-Case]. What I understand from this chart is that a ~500Hz/0.002s pulse approaches almost 0.3oC/W (75% of the 0.4oC/W steady state), which would indicate that the FET actually does have enough time to do substantial thermal cycles at a fair commutation frequency. Of course this is all moot if we just use a worst case scenario as you described using no commutation, which is more representative of the controller's possible operation anyways. We also have to remember that in the real world the highest phase currents will mostly be happening at low speeds too, so this is the really the most stressful time for a controller!


Doctorbass said:
I guess it's the Sil-Pad 1500ST, (8 mil thick)
That one looks pretty good Doc, or check out the hi-flow-650P from the same guys. I've never tried either of them but it's ~30% better than the 1500ST going by their numbers.


Doctorbass said:
The 0.5 mil mylar seem a great idea, but the less thick the material is(0.5 mil), the more it can break with the sharp edges of the mosfet?
All I do is polish them up using very fine sand paper on a flat surface - make sure the whole copper surface has been polished just as John described doing with the heatsink surface. This works out well for me with 1mil kapton, and I don't even use a super-ultra-flat surface as you guys love and worship! ;) I would have to pull out the test numbers, but from memory it made a very good thermal path and didn't cut through the 1 mil kapton.

One setup I used for thermal tests was pretty simple but illustrated quite well the different performances of various thermal materials using real FETs. I just connected a couple of FETs in series using some medium gage silicone wire (not too big nor small), and used mini 2-3S Lipo packs to drive each gate fully On. I then connected this string to a large capacity Lipo pack and an appropriately chosen load resistor for the desired load current along with a switch/breaker to start/stop the test. Each FET can be mounted using a different electrically isolating thermal material, but I made sure the heatspreader / heatsink was *really* thick and massive so that each FET sees a very similar temperature behind them. You can then use the voltage drop of each device under load to evaluate their respective die temps which will often give you a more realistic idea than using an external sensor or IR thermometer. Just make sure the test devices are fairly well matched as far as initial Rds-On at least. I'm not sure about the FET to FET manufacturing variances of the Rds vs Temp slope, but I assume it souldn't be too far appart if the FETs came from the same batch and thus should be doped very similarly. I guess I should've taken photos of this setup to help illustrate, but I'm kinda slack that way I guess. Anyways Doc, If you do this test you will see why I now avoid thick thermal materials of any kind now!

Pat

PS: I second Camlight's suggestion of using either a much fatter heatspreader for each 6 FETs, or dropping them and mounting them directly to the milled heatsink using nice thin kapton/mylar.
 
Hrmm, I was using 1/16th copper plate to solder my banks of 6 fets to. That works out to like 62 mils or something (not used to the term). I hope that it won't be too much an issue with thickness?

I was already planning on lapping the copper bars after having everything soldered to make sure they make proper contact with each other.
 
ZapPat said:
I'm not so sure about this being true even at highish speeds based on the IRFB4110's datasheet, page 5 [Maximum Effective Transient Thermal Impedance, Junction-to-Case]. What I understand from this chart is that a ~500Hz/0.002s pulse approaches almost 0.3oC/W (75% of the 0.4oC/W steady state), which would indicate that the FET actually does have enough time to do substantial thermal cycles at a fair commutation frequency. Of course this is all moot if we just use a worst case scenario as you described using no commutation, which is more representative of the controller's possible operation anyways. We also have to remember that in the real world the highest phase currents will mostly be happening at low speeds too, so this is the really the most stressful time for a controller!

Pat, thanks for reminding me that there will still be some transient effects. This got me thinking, so I spent most of the afternoon doing some modeling for this rather than the grad school work I should've been doing. :)

I started with a pretty big set of assumptions. I looked at the pictures Doc posted a while ago, and did my best to use them to guesstimate some dimensions. Apologies in advance if some of this stuff was posted and I missed it. I assume the FETs are directly connected to the 20 mil strip of copper with basically zero thermal resistance, and I assumed that the strip is just a little wider than the FETs at 0.75". I guesstimated from his pictures that the FETs are about 0.8" on-center spacing, and 3 banks are spaced just a little further apart than that. Underneath the copper is a 1-mil thick kapton insulator (0.12 deg C/m-K). Below that is Doc's aluminum bar, which I assumed to be about 1.25"x0.25", plus the L-section which I assumed to be 0.5"x0.25". The copper tube is in the channel on the back of the Al and there is a good thermal connection between the two. It's really tough to model the water circuit, so I just assumed a resistance of 0.1 C/W from the water to ambient. I'm sure a lot of this will be wrong, but I think the results will still be a useful ballpark guess.

What I did was construct something like a very crude finite elements model. I broke everything down into 0.4"-wide chunks lengthwise along the bar. Conveniently, this is the width of the FETs and at 0.8" spacing exactly one cell fits between FETs. Then I can build an electrical-analog model of the thermal circuit. This let's me account for heat flowing along the heatspreader, and also lengthwise across the Al bar. Based on Pat's comment, I also took into account the transient effects. The heat capacity of the copper and aluminum can be modeled using capacitors. I chose a commutation frequency of 333 Hz, which works out nicely so each bank is active for 1 ms at a time.

I did my own power calculations, assuming 100V, 50% duty cycle, and 400A phase current (200A battery current) as a pretty extreme case. I assumed a hot Rdson of 7.4 mOhm per FET. For the high-side, I came up with 98W of conduction loss and 400W of switching loss (assuming 1 us on/off time and 10 kHz PWM). For the low-side, 197 W conduction and 200W diode loss.

This model ends up predicting a FET junction rise of a little more than 100C at about 500W, so the net junction-to-ambient resistance predicted here is about 0.2 C/W. If we take away the assumed 0.1 C/W for the water cooling, that gives about 0.1 C/W from junction-to-water. These values seem roughly ballpark correct. The very thin heatspreader does seem to be doing some good. The temp in the cells between FETs is about 15 C lower than directly under the FET, so it's far from isothermal but that's still enough to reduce the effective Rth of the kapton layer. One thing that really surprised me is that the model is showing junction temp swings of about 15 C due to the commutation switching, so we should add some margin onto average temp calculations for safety. Also interesting, it takes this model about 2 minutes for the Al heatsink temp to reach steady-state.

I also looked at the thermal effects during the PWM cycle. Using the model in the 4110 datasheet and a pulsed source for the power loss (really high, brief peaks for turn-on and -off, much lower conduction loss), it looks like at 10 kHz it doesn't have much effect. The net thermal impedance was almost exactly the steady-state value of 0.4 C/W.

Here's a picture of the model below.
 
rhitee05 said:
Pat, thanks for reminding me that there will still be some transient effects. This got me thinking, so I spent most of the afternoon doing some modeling for this rather than the grad school work I should've been doing. :)

<snip>

This model ends up predicting a FET junction rise of a little more than 100C at about 500W, so the net junction-to-ambient resistance predicted here is about 0.2 C/W. If we take away the assumed 0.1 C/W for the water cooling, that gives about 0.1 C/W from junction-to-water. These values seem roughly ballpark correct. The very thin heatspreader does seem to be doing some good. The temp in the cells between FETs is about 15 C lower than directly under the FET, so it's far from isothermal but that's still enough to reduce the effective Rth of the kapton layer. One thing that really surprised me is that the model is showing junction temp swings of about 15 C due to the commutation switching, so we should add some margin onto average temp calculations for safety. Also interesting, it takes this model about 2 minutes for the Al heatsink temp to reach steady-state.
Nicely done!
I'm not surprised that the thin spreader has an effect on the overall thermal resistance but I'm worried that the model assumes that the spreader is perfectly bonded to the Al bar over its entire surface. With a spreader that thin, I just don't think that's possible after all the handling it's undergoing (and soldering). I could be wrong. :mrgreen:

Good water cooling setups for CPUs have a sink-ambient resistance of about 0.35C/W. Not sure how to compare that to Doc's bar setup though.

The IRFB4110 has a case-sink thermal resistance of 0.5C/W (assuming proper mounting to a properly greased surface). Did you just add this to the junction-case resistance (0.40C/W) to create one number for the FET resistance (0.9C/W) and then use zero for the case-sink resistance?
 
Doctorbass said:
Thanks John,

I thought that the copper sheet would improve the current sharing and also help the center pin of all fets to get better current share so each of the 6 fet in each bank could heat more equally... no? About the thermal conductibility, i would have used the sil-pad between the copper and the aluminum... so it could fill with good uniformity?

The 0.5 mil mylar seem a great idea, but the less thick the material is(0.5 mil), the more it can break with the sharp edges of the mosfet? I mean i know they are often sharp .. what is your opinion about that?

I have acces for milling job at work.. and i consider that it's a must since i soldered the copper tube on it and it is now not perfectly flat on the back side.

Also, I already lapped aluminum with special lapping tool.. and also ever tried with mirror.. but at work we have a flat marble and some granite ultra flat optical table that are not used... so i have some good references flat ( we used them with some Zygo optical interferometer :wink: . If only i could have the patience to make it flat like a mirror.. but the This is a 'L' bar and not a flat bar.. so the polishing movement can just be made with little circular movements...

Anyway.. i'll do my best.

Oh.. and why Mylar instead of Kapton?

Doc,
Kapton or Mylar will work great for this application, whichever you can find in 0.5mil thickness. Once you get this thin, just about any material will work great. Both of those plastics are very good thermal insulators, but at that thickness it doesn't make much of a difference. :) Hard-coat anodizing is a very effective thermal insulation layer. But, at 0.5mils or thinner, it's a fantastic thing to use as the interface between a FET and the heat sink and has probably the lowest thermal resistance of anything you can use (and it's incredibly tough and has a very high dielectric strength too).

While a sharp point sticking out from a FET can break through a thin insulator, the odds of it being only 0.5mil long, and not a few mils long, is pretty low. Just about any insulator you use will be penetrated once you press down against the MOSFET (when mounting it). I've never had a problem with sharp points/edges though, and no manufacturer ever checks the FETs they use for sharp projections or edges, but you can (as mentioned earlier) just place each FET back-side down on 1200 grit paper, or something similar, and swirl it around a bit to break any edge(s) you feel might be a problem.

The copper sheet would help to reduce "wire" impedance/inductance and therefore help to balance currents better between the FETs and help to hold down any voltage spikes. But the heat spreading will only occur of the copper sheet is perfectly flat all along its length and in 100% contact with the aluminum bar. Otherwise, if the sheet is only held very tightly against the bar underneath the FETs, it's not helping to spread the heat. In fact, the extra layer just increases the overall thermal resistance of the assembly. Even in very high perfomance cooling assemblies, with very high budgets for parts, you won't see the use of a thin spreader. Better to make it a lot thicker or just skip it IMHO.

If you're using the 20mil copper sheet as a conductor, IMHO it's best to just use it as that and keep it out from under the FETs. Makes the entire assembly easier to put together. Your call though...this amazing thing is your project and you should build it the way you feel is best. :D
 
CamLight said:
Nicely done!
I'm not surprised that the thin spreader has an effect on the overall thermal resistance but I'm worried that the model assumes that the spreader is perfectly bonded to the Al bar over its entire surface. With a spreader that thin, I just don't think that's possible after all the handling it's undergoing (and soldering). I could be wrong. :mrgreen:

Good water cooling setups for CPUs have a sink-ambient resistance of about 0.35C/W. Not sure how to compare that to Doc's bar setup though.

The IRFB4110 has a case-sink thermal resistance of 0.5C/W (assuming proper mounting to a properly greased surface). Did you just add this to the junction-case resistance (0.40C/W) to create one number for the FET resistance (0.9C/W) and then use zero for the case-sink resistance?

You're right that the model assumes good contact along the entire surface of the spreader. With Doc's clamping setup I'm sure that will be true underneath the FETs, but the 20 mil copper will be too thin to distribute the pressure evenly between them. I suppose I could tweak the model to reflect this, but it would pretty much be a WAG what the effective Rth of that connection would be. Could be quite high, actually. I do think I'll try running a modified version which assumes a thicker bar, maybe 1/8" copper. Besides making the spreader more effective, I think that will reduce the temp ripple due to commutation, since the copper will have a higher thermal capacitance.

After thinking about it more, I think it's probably best just to delete the water cooling part from the model altogether - it's just hard to make it anywhere near accurate. The model could still be used to predict the backside temp of the sink, where the tube is located. We could probably use one of Doc's experiments to construct an approximate model of the water cooling from there.

I'll admit I totally neglected the case-to-sink resistance. I'm sure with this setup it will be lower than 0.5 C/W, for FETs soldered directly to copper, but it's not going to be zero either. I don't think that'll make a huge difference, though. Each FET dissipates on average 28 W in the model, so an extra 0.2 C/W will only add about 5-6 C to the junction temps.
 
Following up on John's comments, tried out a few variations of the model.

First, I added 0.2 C/W to the FET-to-sink resistance per John's comment. That seems like a reasonable value for a soldered-down FET, unless anyone has a better suggestion. It shows about a 53 deg-C rise from junction-to-water, only a couple of degrees higher as expected.

The next change was to adjust the model to assume poor contact between the copper strip and the kapton in between FETs. To do this I just changed the thermal resistance of the kapton layer to 10 C/W (previously was about 1 C/W for good contact). This increased the peak junction rise to about 60 deg-C. It's pretty clear that the kapton layer is the main limiting factor in the assembly - the difference in temp between the copper and aluminum layers is about 30 deg-C, so the kapton represents about half of the total thermal impedance of the assembly.

The third test was assuming a thicker bar of copper, I used 1/8" here. The model assumes this would be thick enough to spread the pressure out and get good contact across the whole surface. It doesn't actually help as much as I would've thought. The FET rise is now 48 deg-C peak junction-to-water, so it's about 10% better than the 20-mil case. It does decrease the influence of the kapton somewhat, it's now about 1/3 of the total impedance. I expected this would also help reduce the temp ripple, but it didn't - I guess that indicates that the junction-to-sink resistance is mostly responsible for that.
 
A note on Kapton materials: not all Kaptons are the same.

Eg 'Dupont Kapton MT' posseses "3x the thermal conductivity and cut-through strength of standard Kapton® HN".
http://www2.dupont.com/Kapton/en_US/products/MT/index.html

This seems pretty significant as it suggests 3 x 3 = 9x the thermal performance per unit of cut-through strength.

Tape versions are advertised here with samples upon request:
http://www.ipslimited.com/tapes/kapton-tape.htm
http://cshyde.thomasnet.com/viewitems/tapes-with-psa/egories-tapes-with-adhesive-thermally-conductive-2

I bought a small sheet of the film version without adhesive on eBay last year but can't find it on there now...


Also, a note on flat surfaces for polishing FETs, heatsinks ect: a piece of thick sheet glass is most excellent.

- Ben
 
Doctorbass said:
CASE

The first problem to solve was finding the right case for the pcb.

After examinating a little bit the board it is evident that the pcb is too wide and that it is waste of space. that too wide pcb would force us to find really large case and WE DONT NEED THIS :wink:

So the first thing i did was to measure and cut to keep only the usefull width of the PCB.

Then finding the right case that is easy to find and no need for custom made.

And... finally i found it!.. it's Hammond extruded aluminum case that just fit nearly perfect with or need !

PCB dimensions
There was a little but solvable problem about the width of the case. The max width of that Hammond case serie was to allow 100mm pcb. and if cutting only the empty portion of the pcb, that still dont fit. Since i will add strong big cooper bar to the high and low side supply, the pcb trace no more need to be so wide on the high side.. so i can cut it! :wink:

That's what i did to free some width on the pcb.

Then... it slide perfectly into the new case!!!


CAPACITOR

Another little problem here but again that can be corrected easy and btw amelioration the performances of the controller is to replace the actual capacitors.

The original 470uF 200V caps that Lyen sent me are good for low current and high voltage setup.. but not for ultra high current!!

and... their height was too high to fit into the case and also allowing some room under the pcb to install the copper bus bar on the high and low side.

I found a great solution: downsizing the capacitors and choosing lower voltage tolerance since 100V caps have prooven to work great with 24s lipo setup ( 100.8V)

My final choice was the 680uF 100V nippon chemicon low ESR capacitors bought at digikey and shipped in 1 day!

18mm x 35.5mm.... that's exactly 4.5mm lower height and that is exactly what we need !

at 2$ each shipped that's 14$

100V
680uF
27miliohm ESR !!!!
2.5A ripple current.

Here is some pics:


Hello Doctorbass,

i'm new here in this wonderful endless-sphere forum, that catched my attention, while searching for high power controllers for the crystalyte x5 hubmotor. I'm living in Berlin/Germany coming from a bike-messenger background and thinking about two high power ebike projects i like to realize. The first one : very high speed project and second one: very high torque Transportation/Cargo-bike. I read a lot from your different threads and learnt so many things from your profound knowledge, experience, expertise means from your know-how and good skills. Also i'm very much impressed about your pioniering spirit to push forward the ebike-revolution and share ethrything with the comunity. Thank you.

I'm very excited to hear, how are the developing of your 36mosfet controller project is going on, because no news in the thread since april.
I read you did a new speed record with the kelly contoller and planing this sommer to achieve 88mph/142kmh with your watercoolled 36fet Infineon controller. Sounds realy cool. Hope you will be successful.

I really have no skills in elektric/electronic engineering and a very limited electrophysical anderstanding. :roll:

Now i have some questions to you for an better anderstanding and also for developing my projects. :?:

What is the difference between Impedance and ESR ?
Because on the spec.-sheet of your choosen one cap are mentioned 27mohm Impedance and for ESR states - . Because some Companys specify only ESR some only Impedance some both and some nothing.
http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=565-1745-ND

You wrote the Cap has 27mohm ESR but in the spec.-sheet it's specified as 27mohm Impedance. Can you please clarify that.

What is the max. diam. of the caps fitting on the PCB ? On the pictures it looks like the original caps from Lyen (200V, 470µf) Rubycons have 10mm lead spacing snap in case, that means 22mm cap dia. The 100V, 27mohm Chemicons have 18mm dia , 7,5mm lead spacing.
So it looks like that 22mm caps with 10mm snap in case fits on the board. Is this so?

How is the watercooling of the controller going on? I saw your video and different assembling modells in the thread. Any news about progress in design and function? Is it working ? Did you some tests?

Hope to hear from you some exciting news and answers to the questions. :?:

Thanks.

Sharadona
 
sharadona said:
Doctorbass said:
CASE

The first problem to solve was finding the right case for the pcb.

After examinating a little bit the board it is evident that the pcb is too wide and that it is waste of space. that too wide pcb would force us to find really large case and WE DONT NEED THIS :wink:

So the first thing i did was to measure and cut to keep only the usefull width of the PCB.

Then finding the right case that is easy to find and no need for custom made.

And... finally i found it!.. it's Hammond extruded aluminum case that just fit nearly perfect with or need !

PCB dimensions
There was a little but solvable problem about the width of the case. The max width of that Hammond case serie was to allow 100mm pcb. and if cutting only the empty portion of the pcb, that still dont fit. Since i will add strong big cooper bar to the high and low side supply, the pcb trace no more need to be so wide on the high side.. so i can cut it! :wink:

That's what i did to free some width on the pcb.

Then... it slide perfectly into the new case!!!


CAPACITOR

Another little problem here but again that can be corrected easy and btw amelioration the performances of the controller is to replace the actual capacitors.

The original 470uF 200V caps that Lyen sent me are good for low current and high voltage setup.. but not for ultra high current!!

and... their height was too high to fit into the case and also allowing some room under the pcb to install the copper bus bar on the high and low side.

I found a great solution: downsizing the capacitors and choosing lower voltage tolerance since 100V caps have prooven to work great with 24s lipo setup ( 100.8V)

My final choice was the 680uF 100V nippon chemicon low ESR capacitors bought at digikey and shipped in 1 day!

18mm x 35.5mm.... that's exactly 4.5mm lower height and that is exactly what we need !

at 2$ each shipped that's 14$

100V
680uF
27miliohm ESR !!!!
2.5A ripple current.

Here is some pics:


Hello Doctorbass,

i'm new here in this wonderful endless-sphere forum, that catched my attention, while searching for high power controllers for the crystalyte x5 hubmotor. I'm living in Berlin/Germany coming from a bike-messenger background and thinking about two high power ebike projects i like to realize. The first one : very high speed project and second one: very high torque Transportation/Cargo-bike. I read a lot from your different threads and learnt so many things from your profound knowledge, experience, expertise means from your know-how and good skills. Also i'm very much impressed about your pioniering spirit to push forward the ebike-revolution and share ethrything with the comunity. Thank you.

I'm very excited to hear, how are the developing of your 36mosfet controller project is going on, because no news in the thread since april.
I read you did a new speed record with the kelly contoller and planing this sommer to achieve 88mph/142kmh with your watercoolled 36fet Infineon controller. Sounds realy cool. Hope you will be successful.

I really have no skills in elektric/electronic engineering and a very limited electrophysical anderstanding. :roll:

Now i have some questions to you for an better anderstanding and also for developing my projects. :?:

What is the difference between Impedance and ESR ?
Because on the spec.-sheet of your choosen one cap are mentioned 27mohm Impedance and for ESR states - . Because some Companys specify only ESR some only Impedance some both and some nothing.
http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=565-1745-ND

You wrote the Cap has 27mohm ESR but in the spec.-sheet it's specified as 27mohm Impedance. Can you please clarify that.

What is the max. diam. of the caps fitting on the PCB ? On the pictures it looks like the original caps from Lyen (200V, 470µf) Rubycons have 10mm lead spacing snap in case, that means 22mm cap dia. The 100V, 27mohm Chemicons have 18mm dia , 7,5mm lead spacing.
So it looks like that 22mm caps with 10mm snap in case fits on the board. Is this so?

How is the watercooling of the controller going on? I saw your video and different assembling modells in the thread. Any news about progress in design and function? Is it working ? Did you some tests?

Hope to hear from you some exciting news and answers to the questions. :?:

Thanks.

Sharadona


Hello, ESR and impedense can be considered as the same.. normally impedense is specifird Equivalent Serie Resistor at a given frequency.

The 36 fets controller is on hold now due to the not easy to find machine shop to rework my heatsink bar..

I spend alot of time working on my difefrent ebikes.. but sometime i want also to use them when weather is good :wink:

This 36 fets controlelr is not on hold for long time.. i'll continu that soon.

I also think that the kelly might be a better candidate for speed record because it can take up to 136VDC and that at 100vdc i reach 112kmh.

Having a controller of 36 fets of the 4110fets (100V max will not give me more speed.. just better aceleration)



Doc
 
Hello doc,

thanks for the quick answer and news update for the 36 mosfet controller. :D

to this question you didn't answer.

What is the max. diam. of the caps fitting on the PCB ? On the pictures it looks like the original caps from Lyen (200V, 470µf) Rubycons have 10mm lead spacing snap in case, that means 22mm cap dia. The 100V, 27mohm Chemicons have 18mm dia , 7,5mm lead spacing.
So it looks like that 22mm caps with 10mm snap in case fits on the board. Is this so?


sharadona
 
sharadona said:
Hello doc,

thanks for the quick answer and news update for the 36 mosfet controller. :D

to this question you didn't answer.

What is the max. diam. of the caps fitting on the PCB ? On the pictures it looks like the original caps from Lyen (200V, 470µf) Rubycons have 10mm lead spacing snap in case, that means 22mm cap dia. The 100V, 27mohm Chemicons have 18mm dia , 7,5mm lead spacing.
So it looks like that 22mm caps with 10mm snap in case fits on the board. Is this so?


sharadona

Hello, The max actual ( no mod needed) diameter of caps is 18.17mm and 7.5mm lead spacing for the 200V 470uF Rubycon

The max diameter you can ( if you relocate the 0.1uF little red caps) would be 25mm diameter.. but they will be squeezed between the two row of mosfet front facing.

The caps i installed (100V 680uF) are 18.19mm diameter but i preffered to relocate the 104 ( 0.1uF) capacitors under the pcb.

If you drill some new holes.. I can say YES.. the 22mm dia caacitor should fit perfectly :wink:

Doc
 
Hi doctorbass,

i'm little confused, because i thought the 200V 470µf Rubicon caps are dia 22mm, 10mm lead sp. with snap in case.
On the pics of the caps i saw that are caps from MXR series. (your pics, and also from Lyens pics of his 24 mosfet controller) http://endless-sphere.com/forums/viewtopic.php?f=31&t=19719&p=423189#p423189

On Rubycon website datasheet for MXR series i saw the 200V 470µf caps has 22mm dia 40mm height 10mm lead sp. snap in. http://www.dartelektronik.com.pl/pdf/rubycon-serie-mxr.pdf

This MXR caps was on your PCB, but you said they have 18.17mm, 7.5 lead spacing. Confusion :!: :roll:

Are the lead spacing holes on the 36 fet PCB 10mm or 7.5mm ? :?: You said its possible to drill new holes. You did this ? :?:

thanks.

Sharadona
 
sharadona said:
Hi doctorbass,

i'm little confused, because i thought the 200V 470µf Rubicon caps are dia 22mm, 10mm lead sp. with snap in case.
On the pics of the caps i saw that are caps from MXR series. (your pics, and also from Lyens pics of his 24 mosfet controller) http://endless-sphere.com/forums/viewtopic.php?f=31&t=19719&p=423189#p423189

On Rubycon website datasheet for MXR series i saw the 200V 470µf caps has 22mm dia 40mm height 10mm lead sp. snap in. http://www.dartelektronik.com.pl/pdf/rubycon-serie-mxr.pdf

This MXR caps was on your PCB, but you said they have 18.17mm, 7.5 lead spacing. Confusion :!: :roll:

Are the lead spacing holes on the 36 fet PCB 10mm or 7.5mm ? :?: You said its possible to drill new holes. You did this ? :?:

thanks.

Sharadona

I see your confusion :D ... I did NOT installed the XMR capacitor... its the caps that WAS ALREADY on the pcb!!

I removed the XMR and installed the brown one!.. if you read carefully you will see 680uf 100V :wink:

IT'S THE KZE serie from nippon chemicon that i installed.

And maybe the 24 fets and the 36 fets board are not too similar... the real spacing of the capacitor HOLES ON THE PCB IS 7.5mm MEASURED.
 
ok. i knew you installed the better 100V KZE caps and the Rubycons was originally on the Board.

But my confusion is still there.

If the Rubycons are 18.17mm like you said everything is clear and it is like you said that for bigger caps new holes shood be drilled. And it means its wrong Info from the spec.sheet. or its not MXR serie like it looks.

But if The Rubycon MXR are 22mm , 10mm snap in lead sp. like the spec.-sheet said, that would mean that they fitted in the 7.5 mm holes without modification. That would be cool.

Therefore i ask , if you are sure that the Rubycon caps hase 18.17mm. Maybe you can please remeasure the old Rubycons ?

Sorry for troubles.


Sharadona
 
Please go at the first page of this thread

you'll see that i made a comparaison between the rubicon and the nippon capacitors in the 5th and 6th picture!

They are the same diameter and pin spacing
 
Hi Doc,

ethrything is now clear. Thanks.
I did some research for components what i like to use for a highspeed project that i'm planing. Want to know, what you think about of the comps, thoughts and ideas.
Maybe some will fit for your 36 fet controller on high speed mode 150V (144V). And maybe your next effort for a speedrecord test will be with this controller.


OK. I found very interesting cases/enclosures made from extruted Aluminium that are bigger in height , even ajustable in height, available in silver or black. That cases allows using caps with more height for instance 45mm cap.-height.
Also for the watercooling system would be more room or space. Most properly cover plate for heatsink application is available.

http://www.enclosuresandcases.com/sbspecpage.html
http://www.enclosuresandcases.com/fspecpage.html


to the fets :

of course the good irfb 4115 http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=IRFB4115PBF-ND 11mohm Rds on 104A
the infineons IPP075N15N3 G looks also interesting :!: http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=IPP075N15N3%20G-ND only 7.5mohm Rds on 100A but very expencive

the caps: all are -40 to 105°C

miniature 18mm Capacitance Ripple current@ 120Hz

160V Nichicon cy-serie 18*35mm 470µf 1.92A http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=UCY2C471MHD-ND
18*40mm 560µf 2.13A http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=UCY2C561MHD-ND

200V Nichicon cy-serie 18*40mm 390µf 1.78A http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=UCY2D391MHD-ND
//www.nichicon.co.jp/english/products/alm_mini/list_f.htm

Un. Chemicon KXJ-serie 18*35mm 390µf 1.565A
18*40mm 470µf 1.745A http://www.chemi-con.com/components/com_lcatalog/uploaded/7/7/7/5228327064d9b14b445893.pdf

Rubycon TXW-serie (18*35mm 390µf 1.43A)
(18*40mm 470µf 1.58A)
18*45mm 560µf 1.77A http://www.rubycon.co.jp/de/catalog/e_pdfs/aluminum/e_TXW.pdf


but for much more performance :!: :?: i would prefer 22mm caps. You said, by drilling new holes they will fit. I read in a thread that you used the 22mm rubycon cap(40mm 1000µf 2.2A i think) on your 18fet controller for your speed record on your Giant DH Comp. So, it looks like you allready drilled some new holes for 22mm caps in the board for the 18fet controller. Is it so? :?:

large can 22mm caps:

160V Rubycon MXG-serie 40mm 1000µf 2.2A 3000h http://www.rubycon.co.jp/de/catalog/capacitors.html
Panasonic TS-HC 45mm 1200µf 2.3A 2000h http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=P13953-ND 166mohm ESR!

200V Nichicon GG-serie 40mm 820µf 1.75A 2000h http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=LGG2D821MELZ40-ND http://www.nichicon.co.jp/english/products/alm_larg/list_f.htm
Nichicon GG-serie 45mm 1000µf 2.04A 2000h http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=LGG2D102MELZ45-ND
Nichicon GW-serie 45mm 820µf 2.65A 3000h http://search.digikey.com/scripts/DkSearch/dksus.dll?Detail&name=LGW2D821MELZ45-ND http://www.nichicon.co.jp/english/products/alm_larg/list_f.htm
(very high ripple current and most properly the lowest ESR)

What do you think is the best combination fet-cap for the best high speed performance? :?:

Would be happy if you can use some of the comps. for your project. I'm a great fan of your projects.


Sharadona
 
Doc.. can the boards still be purchased today? i know - this is an old thread.. but the cromotor is coming out.. i thought i would bump it :)
 
Doc.. can the boards still be purchased today? i know - this is an old thread.. but the cromotor is coming out.. i thought i would bump it

Neptronix, if you are looking for an out of the box controller for the CroMotor, the stock Lyen 24 Fet "Muscular" controller http://www.endless-sphere.com/forums/viewtopic.php?f=31&t=19612&p=364552&hilit=lyen+muscular#p364552, will easily have enough balls for the Cromotor (particularly the current lower KV motor that went out as the last batch). Even Zombiess is running (an albeit heavily modified) 18 Fet for his, and running 80-100Amps. I don't know what it is you are attempting to feed into it, but the Lyen 24 Fet stock is rated to 10kw, so I think you will have motor heatsoak problems way before the 24 Fet doesn't have enough enough balls for you.

I have no doubt that the 5403 I have on order will suffer heat problems way before my Lyen 24 Fet blows.

My 5303 melted a 12 Fet, exploded an 18 Fet, but has never been able to get my 24 Fet even slightly warm, all limitations on the use of that motor for me were related to motor heatsoak. Having said that I am only running 20S, and have never run it over 80amps battery, I don't know what it was you were going to attempt to inject into the Cromotor, if you are attempting 16-20kw lunacy like the good Doctor, then by all means a 36 Fet might be a sensible avenue to pursue......
 
Funny you should ask that...I was on the search for a controller for my new build.
I was thinking Kelly till David on ES Facebook page pointed in the direction of this threadLukes thread about them.

So the search was on again. and David mentioned this thread and my interest was again spiked.
OK, I am not going for a cromotor, but a 5404..although I have asked for a 5403 wind and plan on a 17 inch motorcycle rim which I think is going to give me a rolling diameter of around 20-22 inch depending on the tyre.

I can't go for speed ( I live on an island where all island, all road speed limits for all vehicles is 40 mph) ..so All out max of 55 to 60mph is al I need. But I want acceleration. gotta out perform the kids on the 125 m/cycles :twisted:

Hoping to go up to the 100 volt plus range...if appropriate for my needs..32-34 series?

Am i correct in thinking that for a specific voltage, a 5xxx series motor in 05 wind will be more torquey and less top end than a 5404 ? or have I got the basics wrong (again)


So any advice to achieve my aims greatly appreciated...
 
Back
Top