[ga2500ev]

I wanted to restart the discussion on non PWM controllers. PWM controllers have been discussed to death here. However, because they stress all their electronics at full pack voltage at all times, the construction of one DIY seems to be a significant challenge.

Non PWM controllers, including contactor controllers, rectactor, controllers, and the BatPack controller seem to offer some advantages over PWM controllers:

1) They have higher efficiency due to less switching losses.

2) They function in limp mode even if part of the controller fails. This gives a higher measure of reliability.

3) Subcomponents have less stress, so can be lower voltage and offer higher efficiency. Plus they are cheaper.


So I wanted to talk about there types of controllers and ask a few questions. My goal is to design a controller that can function from 96-144V.

The Contactor/Rectactor

The basic design was offered by Lee Hart in an EVDL post. The basic two battery design looks the image loaded below.

This is a serial/parallel controller. First S1A/S1B are turned on together to connect the batteries in parallel. Then S2 bypasses the current limiting resistor which bumps the speed. Finally the S1 switches are turned off and S3 turned on which connects the batteries in series, doubling the voltage to the motor.

The S1 switches, when replaced with diodes becomes a rectactor. At low speeds there is a diode drop of voltage across each battery. As the S3 switch switches in, the diodes are bypassed providing full power to the power. Also there is a need for a main contactor because with the diodes the minimum voltage is always routed to the motor.

A more modern idea is the BatPack. You can find it here:

http://www.redrok.com/ev.htm

I've talked about it in the past. The idea is to connect BatPack modules (shown below) serially. Each module can be a different voltage (unlike the contactor/rectactor (CR) designs). The controller powers a subset of the modules, by switching on the SP switch, to power the vehicle, while others are bypassed, using the SB switch. As with the rectactor, the SB switch can be substituted with a diode and its corresponding diode drop.

Now the BatPack, unlike the CR designs, drains the batteries in the subpack asymmetrically. In a CR design, all the batteries are used at every step. But with the BatPack, are dropped out which means that no power is being drawn. The BP uses battery management to monitor the voltage of each pack and to switch in stronger packs while dropping out weaker packs for a particular voltage. So for example if we have a BatPack with 2 12V modules and 2 24V modules and the driver is calling for 36V, then the controller would select the strongest 12V pack and the strongest 24V pack and turn them on. As long as they are the strongest voltage packs they remain selected for that voltage. But if their voltage drops below the corresponding unused pack, then that pack is switched in.

Now to my questions. I'm really interested in the BatPack. The design makes sense. My first question is that if the SP switch is implemented with MOSFETS then where can you draw the gate voltage required to turn on the MOSFETs? You need the gate voltage to be above the source, which is connected to the positive terminal for the battery in the module. I presume that a charge pump would need to be implemented to generate the gate voltage?

The second question is about the bypass switch. While using a diode is an easy solution, the added diode drops will burn off quite a bit of energy as heat. What would happen if you replaced that switch with MOSFETs? Would those MOSFETS have to support the voltage of the entire pack? Or just the voltage of the battery that is bypassed? Or is it related to the voltages above/below that switch?

Any thoughts welcome.

ga2500ev

[Qer]

I have a serious problem combining these two statements describing the same construction:

Originally Posted by ga2500ev

1) They have higher efficiency due to less switching losses.

Originally Posted by ga2500ev

Then S2 bypasses the current limiting resistor which bumps the speed.

Switching losses are usually just a few percent (maybe as low as 1-2% depending on the construction), current limiting resistors usually introduce losses of tens of percents. If you introduce any kind of resistance (no matter if it's in the form of resistors, excessive cabling, resistance in the contactors) your efficiency will drop like a stone in vacuum. On Saturn.

I also doubt this part:

Originally Posted by ga2500ev

2) They function in limp mode even if part of the controller fails. This gives a higher measure of reliability.

I worked a while for the Swedish telephone company (back in the days where there were ONE telephone company owned by the government) and one of my tasks were doing maintenance on the relays (this was back in the days where most telephone stations still used relays rather than computers) because relays, or contactors, need maintenance if they're heavily used. The contactor in an ordinary EV doesn't switch very often during a year when it's only activated once per trip but if you have a bank of contactors that regulates throttle they will probably need regular maintenance. So I think that you will find out that the reliability actually will be lower, at least compared to a good silicon based controller like the Zilla.

But it will definitely make a very interesting sound, much like the trams I rode when I was a kid. They clattered a LOT when the driver changed acceleration where modern trams are pretty boring since they're almost silent.

Just make sure you're adding enough fuses to avoid an accident if, for example, a contactor hangs (it does happen, especially when they start to get worn down). Shorting a battery pack, or parts of a battery pack, is not something you want to risk...

[ga2500ev]

Originally Posted by Qer

I have a serious problem combining these two statements describing the same construction:





Switching losses are usually just a few percent (maybe as low as 1-2% depending on the construction), current limiting resistors usually introduce losses of tens of percents. If you introduce any kind of resistance (no matter if it's in the form of resistors, excessive cabling, resistance in the contactors) your efficiency will drop like a stone in vacuum. On Saturn.

That resistor is in place to limit the initial torque at low voltage so the vehicle doesn't shoot off like it's been loaded in a cannon. it's only for the first few seconds from a standing start.

OTOH at 16 KHz there are 2 switch transistions every 62.5 uS. Even if the switch time is only 1 uS per switch then that's a 3.2% loss all the time.

Also if you implement with semiconductor switches, then you can PWM those switches to limit the torque initially then have them switch permanently once the vehicle is moving.

I also doubt this part:

I worked a while for the Swedish telephone company (back in the days where there were ONE telephone company owned by the government) and one of my tasks were doing maintenance on the relays (this was back in the days where most telephone stations still used relays rather than computers) because relays, or contactors, need maintenance if they're heavily used. The contactor in an ordinary EV doesn't switch very often during a year when it's only activated once per trip but if you have a bank of contactors that regulates throttle they will probably need regular maintenance. So I think that you will find out that the reliability actually will be lower, at least compared to a good silicon based controller like the Zilla.

Qer, I think you missed my point. At the end I was asking how can you implement the structure of these types of controllers using semiconductor switches? So you use MOSFETS instead of actual contactors.

It's the structure of the controller that I want to keep, not the actual parts. I don't want to use actual contactors because frankly they cost too much, are too noisy, and as you pointed out, require maintenance.

Any controller structure is going to require switches. The question is how do you organize them and how fast do you switch them?

Go back to the rectactor. Imagine that S2 and S3 are MOSFETS and S1 diodes. So there are no contactors. Now since S2 (which fundamentally is the PWM switch in a PWM controller) is a semiconductor switch, it can be PWMed at low speeds then locked in once the vehicle has gotten up to speed.

Same with the BatPack modules. They are already switched at 100 Hz. Turn on one 12V pack at half power and you have a 6V source to the motor.

I certainly do not want actual contactors. However, I am very interested in figuring out how to replace those contactors and diodes with MOSFET switches.

But it will definitely make a very interesting sound, much like the trams I rode when I was a kid. They clattered a LOT when the driver changed acceleration where modern trams are pretty boring since they're almost silent.

Just make sure you're adding enough fuses to avoid an accident if, for example, a contactor hangs (it does happen, especially when they start to get worn down). Shorting a battery pack, or parts of a battery pack, is not something you want to risk...

Again I'm interested in the structure, not the components.

ga2500ev

[Qer]

Originally Posted by ga2500ev

OTOH at 16 KHz there are 2 switch transistions every 62.5 uS. Even if the switch time is only 1 uS per switch then that's a 3.2% loss all the time.

Well, if you're using MOSFET's as you suggest further down there will be a lot of Ron's in the chain which will result in rather substantial losses as well. I'm not convinced it's an improvement. Also, I believe our numbers for our prototype is better than 3.2% losses, but I'll leave that to Tesseract to sort out if he likes.

Originally Posted by ga2500ev

Also if you implement with semiconductor switches, then you can PWM those switches to limit the torque initially then have them switch permanently once the vehicle is moving.

But then you will have to implement both a PWM-controller and a PWM-free controller, ending up with twice the problems...

Originally Posted by ga2500ev

Qer, I think you missed my point. At the end I was asking how can you implement the structure of these types of controllers using semiconductor switches? So you use MOSFETS instead of actual contactors.

Well, it's doable but considering the amount of contactors/MOSFETs/whatever you'll need, it's not going to be cheap. The simplest approach is probably to use solid state relays but they don't come cheap. Especially not if they need to handle DC. How many steps do you want the controller to handle? If it's enough with 3 power steps (1/4, 1/2 and full pack voltage) you still need 9 switches that each can handle max current.

Looking at Digikey the strongest solid state relay for DC I can find handles max 100 Ampere and 60 Volt (which, to say the least, is a bit on the low side) and they cost $120. Each. They have a MOSFET that can handle 200 Volt 580 Ampere and it "only" cost $145, but you still need at least 9 of them. You can probably find much cheaper on E-bay, but in case you want a little more than three speeds the amount of MOSFET's will increase drastically and thus also the cost.

Then there's another problem involved, if you use silicon you need to limit the current and make damn sure you don't exceed the rating. If you do it by starting to PWM when things go rough you will enter that twilight zone where spikes and heat dissipation blow things up for you and if you do it by switching down to a lower voltage you will never be able to get maximum current out of your pack and the performance will suffer.

But it's definitely doable, just not very practical. However, to answer some of your questions:

Originally Posted by ga2500ev

My first question is that if the SP switch is implemented with MOSFETS then where can you draw the gate voltage required to turn on the MOSFETs?

For S3 and S1b you can take the power from the battery/batteries they switch. For S1a you'd need to get a voltage that's sufficiently higher than the local batteries so you probably need some kind of local DC/DC too. To isolate the MOSFET's from the rest of the electronics you can connect the gates of the MOSFET's through opto-couplers, one for S1a and S1b and one for S3.

Originally Posted by ga2500ev

The second question is about the bypass switch. While using a diode is an easy solution, the added diode drops will burn off quite a bit of energy as heat. What would happen if you replaced that switch with MOSFETs? Would those MOSFETS have to support the voltage of the entire pack?

I'm not quite sure, but I think the MOSFET will have to survive HIGHER voltage than the pack voltage thanks to back-EMF since you will have to make DAMN sure SP is open before SB closes and during that time back-EMF will skyrocket to several hundred Volts, maybe even beyond a thousand Volt. It's probably easier to keep the diode because at the currents we're talking about I don't think the losses will be that big due to the diodes Vf compared to the MOSFET's Ron.

But, seriously, this is not going to be neither a good, nor cheap controller. There's a reason more or less all motor controllers out there (not just in EV's but in process industry etc) runs PWM. It's the most practical solution and done right it's very reliable. That some controllers has blown doesn't mean the concept in itself is faulty...

[Anaerin]

Originally Posted by Qer

But then you will have to implement both a PWM-controller and a PWM-free controller, ending up with twice the problems...

I could be wrong, but isn't this how the Curtis controller works? It PWMs at low power requests (To vary the voltage/torque), then WOTs the PWM circuitry to full pack voltage and varies the amperage to change speed. Hence the "Violin" sound at crawling speeds.

Or have I missed the mark here?

[Qer]

Originally Posted by Anaerin

Or have I missed the mark here?

Slightly.

The Curtis usually PWM at 15 kHz but at start it PWM's at 1.5 kHz since there were problems with older Curtis controller blowing up if there were a major current rush. Or something like that, the lower frequency that makes the squealing is a fix on some kind of problem they had before anyway.

So it's always PWM'ing, just at different frequencies.

[Tesseract]

This is not a completely bad idea but realize that it gives very granular control of the wrong parameter, too (you ideally want to control motor amps, not motor volts - motor volts will work but it will give you that jerky "golf cart" feeling).

Originally Posted by ga2500ev

...OTOH at 16 KHz there are 2 switch transistions every 62.5 uS. Even if the switch time is only 1 uS per switch then that's a 3.2% loss all the time.

1uS per transition in this case would be exceptionally sloppy. The rule of thumb is that the ideal total switching time should be around 0.5-1% of the period because that strikes a good balance between overshoot/ringing and switching losses. Also, one big advantage if the load is inductive (ie - a motor) is that the voltage and current hardly overlap at switch turn-on.

FWIW, the switching time per transition on my controller protoype is 150nS at 680A of load current, or a very respectable 0.48% total.

Originally Posted by ga2500ev

Also if you implement with semiconductor switches, then you can PWM those switches to limit the torque initially then have them switch permanently once the vehicle is moving.

If PWM will be used at some point then why bother with the rest of the setup?

Originally Posted by ga2500ev

Qer, I think you missed my point. At the end I was asking how can you implement the structure of these types of controllers using semiconductor switches? So you use MOSFETS instead of actual contactors.

It's the structure of the controller that I want to keep, not the actual parts. I don't want to use actual contactors because frankly they cost too much, are too noisy, and as you pointed out, require maintenance.

No problem with this, but realize that each bank of MOSFETs used to replace a contactor will need to have isolated gate drive and while they do not need to be switched quickly, the timing of the turn off of one "contactor" and the turn-on of another needs to be carefully coordinated. I suspect that in the end your parts cost will be the same or more as a goood PWM chopper design but with much worse throttle response to show for it. Not saying it can't be done or that it isn't worth thinking about, just that it does not pass muster with me on first glance.

[ga2500ev]

Originally Posted by Qer

Well, if you're using MOSFET's as you suggest further down there will be a lot of Ron's in the chain which will result in rather substantial losses as well. I'm not convinced it's an improvement. Also, I believe our numbers for our prototype is better than 3.2% losses, but I'll leave that to Tesseract to sort out if he likes.

Take a read on the BatPack numbers for Ron. Specifically:

I made a leap of logic when I discovered that the on resistance of MOSFETs increased with the BVds to the power of 2.8. This factor approximately was true for all venders. Ok, so with the use of 500 Volt MOSFET used in the 250 Volt switcher vs. the 250 Volt MOSFETs used in the 125 Volt switching controller I would need about 7 times as many transistors connected in parallel.

If these number are even close to true, then that means that sums of the Rons of the lower voltage MOSFETS would be much less than the Ron for the PWM controller.

But then you will have to implement both a PWM-controller and a PWM-free controller, ending up with twice the problems...

The difference is putting your eggs in a single high voltage basket as opposed to multiple lower voltage ones. with significant differences in switching speed.

Also the pseudo PWM I'm talking about is nothing more than adjustable current limiting, of which overcurrent protection is going to be required anyway. We all know that torque is directly related to current.

Paul Homes made an interesting observation in the development of his open source PWM controller. You can find the thread here:

http://ecomodder.com/forum/showthrea...ller-6404.html

He observed that you can get smooth acceleration simply by limiting the percentage of current to the percentage of throttle. So if the throttle is at 10% of max, limit current to 10% of max.

The same can be done for a contactor/batpack type controller at low speeds. Of course the obvious way to pull this off with MOSFET switches is to PWM them during that period. Once reach 100% of the lowest battery level valued subpack, you can disable the PWM and just start switching normally.

Well, it's doable but considering the amount of contactors/MOSFETs/whatever you'll need, it's not going to be cheap. The simplest approach is probably to use solid state relays but they don't come cheap. Especially not if they need to handle DC. How many steps do you want the controller to handle? If it's enough with 3 power steps (1/4, 1/2 and full pack voltage) you still need 9 switches that each can handle max current.

The number of switches depends on the configuration. In the rectactor config you'd need 7 switches to switch 4 modules. In the BatPack config you'd only need 3 modules and therefore 3 switches.

Now we're talking! Again take a look at BatPack argument. The lower valued MOSFETS have a lower Ron. So each can carry more current at their rated voltage.

I'd like to see some holes poked into that argument. Is it true that the RDSon jumps 7 times when you double the max voltage of the MOSFET? Let me go see for myself. I'm going to look up the RDSon for a 50V, 100V, and 200V MOSFET in the same family. Be right back.

Took a minute. Doesn't seem to hold true. I took a look at International Rectifier HexFETs on digikey. Only took items that were actually in stock. Here's what I found:

irfp064npbf 55V 8 mOhms RDSon 59A $2.82
irfp4410zpbf 100V 7.2 mOhm 97A $3.70
IRFSL4227PBF 200V 26 mOhms 62A $4.14

While the 200V part has a much higher RDSon, the 55V and 100V parts are virtually identical.

Looking at Digikey the strongest solid state relay for DC I can find handles max 100 Ampere and 60 Volt (which, to say the least, is a bit on the low side) and they cost $120. Each. They have a MOSFET that can handle 200 Volt 580 Ampere and it "only" cost $145, but you still need at least 9 of them. You can probably find much cheaper on E-bay, but in case you want a little more than three speeds the amount of MOSFET's will increase drastically and thus also the cost.

But you don't need 200V mosfets. That's the whole point. You can do the whole thing wth 55V or 100V mosfets with the prices I outlined above. BTW the item you referenced isn't in stock. That makes it made of unobtanium. Actually $145 for a 200V 580A MOSFET module isn't too bad a price.

But back to the point. If we took the 100V parts and derated to half max power, then for 400A we'd need 400A/50A = 8 parts per switch. In the batpack config with 3 switches that would be 24 parts. Digikey breaks the price to $1.95 at 10 parts. So the total cost would be about $50 for the switches.

Then there's another problem involved, if you use silicon you need to limit the current and make damn sure you don't exceed the rating. If you do it by starting to PWM when things go rough you will enter that twilight zone where spikes and heat dissipation blow things up for you and if you do it by switching down to a lower voltage you will never be able to get maximum current out of your pack and the performance will suffer.

overcurrent is solved the same way as a PWM controller: You PWM all the switches. You don't want to lower the voltage because then that raises the current required on the remaining switches. So you cut off the power until the current drops and turn on all the switches again.

But that's an overcurrent condition and shouldn't happen under normal circumstances.

But it's definitely doable, just not very practical. However, to answer some of your questions:



For S3 and S1b you can take the power from the battery/batteries they switch. For S1a you'd need to get a voltage that's sufficiently higher than the local batteries so you probably need some kind of local DC/DC too. To isolate the MOSFET's from the rest of the electronics you can connect the gates of the MOSFET's through opto-couplers, one for S1a and S1b and one for S3.



I'm not quite sure, but I think the MOSFET will have to survive HIGHER voltage than the pack voltage thanks to back-EMF since you will have to make DAMN sure SP is open before SB closes and during that time back-EMF will skyrocket to several hundred Volts, maybe even beyond a thousand Volt. It's probably easier to keep the diode because at the currents we're talking about I don't think the losses will be that big due to the diodes Vf compared to the MOSFET's Ron.

But, seriously, this is not going to be neither a good, nor cheap controller. There's a reason more or less all motor controllers out there (not just in EV's but in process industry etc) runs PWM. It's the most practical solution and done right it's very reliable. That some controllers has blown doesn't mean the concept in itself is faulty...

All the discussion that I've seen here makes PWM seem undoable. And it's all related to trying to switch the full pack voltage at 100s of amps at 15 kHz+.

Qer, take your focus away from the rectactor for the sake of discussion and focus on the BatPack. Imagine a 144V 500A controller that consist of 6 subpacks: 2 12V, 2 24V, and 2 36V packs stacked serially. It delivers voltages in 12V increments from 0-144V giving 13 steps.

Now presuming that the you implement both switches using MOSFETs you'd need 20 per module. 120 MOSFETS. At $1.52 each that would run about $170.

Of course you still need the freewheeling diode to protect the electronics.

ga2500ev

[ga2500ev]

Originally Posted by Tesseract

This is not a completely bad idea but realize that it gives very granular control of the wrong parameter, too (you ideally want to control motor amps, not motor volts - motor volts will work but it will give you that jerky "golf cart" feeling).

I'm aware. I can live with the a reliable jerky golf cart feeling.

1uS per transition in this case would be exceptionally sloppy. The rule of thumb is that the ideal total switching time should be around 0.5-1% of the period because that strikes a good balance between overshoot/ringing and switching losses. Also, one big advantage if the load is inductive (ie - a motor) is that the voltage and current hardly overlap at switch turn-on.

Good to know.

FWIW, the switching time per transition on my controller protoype is 150nS at 680A of load current, or a very respectable 0.48% total.

Are you have any overshoot issues?

If PWM will be used at some point then why bother with the rest of the setup?

PWM is going to have to involved because of the semiconductors. With contactors the amps are not as big an issue. With MOSFETS you overamp them and they smoke.

So you are going to have to have overcurrent control which means turning the switches on and off, a form of PWM.

You'll get the same effect if you adjust the current limit at low throttle positions.

But once you get to speed, the switching is minimal.

No problem with this, but realize that each bank of MOSFETs used to replace a contactor will need to have isolated gate drive and while they do not need to be switched quickly, the timing of the turn off of one "contactor" and the turn-on of another needs to be carefully coordinated. I suspect that in the end your parts cost will be the same or more as a goood PWM chopper design but with much worse throttle response to show for it. Not saying it can't be done or that it isn't worth thinking about, just that it does not pass muster with me on first glance.

That's the kind of feedback I'm looking for. But my primary criteria are DIYability and reliability. I'm pretty sure I can stack a 12V, 24V and 36V module together and get a 6 speed test setup that switches at 100 Hz. OTOH I'm almost certain that this guy Paul over at ecomodders who is trying to put together a DIY PWM is going to have his project blow up.

http://ecomodder.com/forum/showthrea...ller-6404.html

There's a vast difference between switching at 4, 8 or, 16 kHz and switching at 100 Hz or less. BTW a $5 microcontroller would have no problem sequencing switching.

Every thread that I've read here on PWM makes it clear that there are nuances that makes it very difficult to DIY.

I need a controller that works, that I can build myself, and that won't cost a mint to build. In all the time I'm been on the forum, no PWM controller meets that standard.

Hence my discussion on the alternatives.

ga2500ev

[Qer]

This mad idea of yours has been nagging in the back of my head for a few days, and I have some thoughts that you might like (or not):

• Skip the MOSFETs. Let's say you have 4 groups of batteries you can swap around with the MOSFETs, that means that if you go full throttle and run at 120 Volt, 500 Ampere (numbers out of a hat just as an example) and you have top notch MOSFETs with Ron of only 5 mOhm that still means that you will lose 500 Ampere * 5 mOhm * 3 MOSFETs = 7.5 Volt. That's 3.75 kW heat you have to cool and it means your efficiency is down to 94%!
• Don't PWM. If you PWM you'll have the same problems as any other PWM-controller plus additional problems as well. If you really want to do a sledge hammer and crowbar controller, do as they did before the silicon era and use a resistor for soft start if you can't switch down the Voltage enough.

My suggestion is that you use contactors that can handle, say, 1000 Amps as long as they don't have to make/break the current (probaby cheaper than contactors that can make/break all that current) for reorganizing the battery pack and then you use one bad-ass contactor that CAN make/break serious current as the major on/off-switch. If you use a micro controller (or maybe build some relay logic to match the contactors ) that can read the contactor position by the help of a monitoring switch on each (sorry, don't know what they're called in english) then a switching sequence would look like this:

• Break master contactor
• Reorganize the battery pack connections
• Wait until all contactors are secured in the right position
• Wait a little longer to make sure there's no contact bounces that can arc and weld
• Switch on master contactor

The reorganization of the pack has to be done carefully (and possibly in several sequences to avoid any risks) to make sure that a slow, or stuck, contactor doesn't result in for example two parallel connected parts of the pack gets connected with two serial connected parts. I'd recommend fuses for every stand alone part of the pack to avoid big badabooms...

I'm still not convinced this is a good idea or that it'll be efficient or cheap, but it will definitely be a controller that almost anyone can build, maintain and repair. However, all those current spikes will probably bring havoc with your peukert effect and ruin your range...

So, that's how I think a DIY-or-bust electro-mechanical controller should be built. Don't forget to make sure the sound comes through loud and clear if you ever upload a video to YouTube! It's definitely going to be more impressive sounding than the Curtis squeal.