Hi everyone!

I'm working on a project for home for learning and fun. I am using peltiers for cooling a water reservoir, I have most everything finished up, but realized that my control circuit could use a bit of help!

The attached schematic is incorrect in that I realized the IRFZ44N was not a Logic level FET and swapped Vcc of the driver to 12v. There is where my troubles started...
[IMG=1000,1000]http://i.imgur.com/HJlD4MQ.jpg/width[/IMG]
When the FET switches there is a 40-50v ringing on the load side and a 20v ringing on the source side, the ringing popped my both my FET and my driver, oops!

A quick correction that I am making is getting a logic level N-MOSFET and removing the 12v from the driver. I will move Vcc of the driver (back) to the 5v line of the power supply to hopefully protect it a bit more!

My question is, could someone help me reduce the voltage spikes on the ringing? It will be next week sometime before the new FET's show up (with more drivers....), but I was hoping to have a plan instead of just throwing them in and hoping for the best, I am not sure where to go from here to reduce the spikes, especially with such a large load rating.

The Schottky is a 45v 20a.
PWM is 31Khz

The load should be just under 20 amps peak current at 100% duty cycle.

I appreciate any help!

Edit: And if anyone can point out how to reduce the size of the monstrous image I posted so it's not so.... BIG, that would be great too

 

Since I seem to have broken the first post... It won't allow me to edit it anymore and I did not want to leave a huge image in the middle of the text, here is try #2. Sorry about the double post.

 

Have you considered using a snubber?

From the controller maker's point of view:
https://www.ti.com/ww/en/analog/power_management/snubber_circuit_design.html

From the capacitor maker's point of view:
www.cde.com/resources/catalogs/igbtAPPguide.pdf

 

Why did you install inductors?

 

The 0.47 inductor was an attempt to filter the ringing back to the PSU. It worked when I was not fully switching the MOSFET... (see above about thinking I had a logic level, oops) It reduced a 1v p-p to around 250mV p-p. Once I fully switched the FET though, yeah, not so much.

The second inductor is to help smooth the voltage from PWM to a nice linear output. I was going to use straight PWM but peltiers run more efficiently at lower voltages (amperages), so I would lose something like 40% efficiency (rough guess) at 50% duty cycle with a 12v pulse vs a linear 6v.

I just started reading up on snubbers this evening. Some of it is still going right over my head, I'll keep trying to soak it in, maybe something will click and I'll understand

 

How about moving your Shottky down, across the mosfet?
I also don't think you need to switch a Peltier at 30KHz to get it to cool things.
There is also the idea that electrolytic capacitors don't work well at high frequencies. Try some 0.1 uf ceramics in parallel with them.

 

The higher switching frequency was so that I could get a smooth voltage. I'm by no means an engineer though! Would the higher frequency be causing a larger spike on switching? I could reduce the frequency and increase the electrolytic size to keep the ripple down, with a ceramic in parallel. Is the ceramic to fill the gap so to speak, why the larger cap catches up?

What does moving the schottky across the FET do in this case? I guess I'm not seeing it (which is why I'm here and not out playing with the circuit by myself!)

 

I'm not seriously an expert on this, but...
What do mean..."smooth voltage"? A Peltier doesn't care about that, and using PWM in the first place is the only reason the voltage isn't smooth, and using inductors is the main reason you have spikes. What are you cooling that needs the cooling to stop in a 30,000th of a second? Most coolers I know run for several minutes, then shut off for several minutes.

Fast switching is the cause of inductive spikes. Energy in the inductor when you switch it off = 1/2 L(I squared) and Energy = 1/2 C (V squared). This says that the energy you are trying to stop by slapping the mosfet off has to go somewhere, and it goes into the capacitors. 45 uh @ 20 amps is 0.009 energy somethings. 0.009 = 1/2 (1000 uf) V squared so your spike should be 4.24 volts on top of the existing 12 volts...but it isn't. That's because electrolytic capacitors don't work very well at near Megahertz speeds. Yes, the ceramic capacitors fill the gap until the electrolytic capacitor catches up.

Moving the Shottky down to the mosfet will protect the mosfet from voltages higher than 45 volts, if it's quick enough to work in a few microseconds (I think it is). But why create the spikes in the first place? Why run a DC device at 30KHz?

 

Peltiers do care about 'smooth' voltage though! Their cooling efficiency is a direct correlation of amperage, I'm probably not the best person to explain this but I'll try a bit.

The COP(efficiency) of a peltier goes way up the lower you run the voltage/amperage, and the heat you have to remove goes way down. If you run it straight PWM the COP is still really really low and the heat being generated is high, you lose a ton of efficiency because you are still pulsing it at a COP of around 0.3.
Can you do it? Sure, just oversize the peltiers and get larger heat sinks. But you would get better results with a linear voltage where your COP is a 1 or higher, that's what I'm shooting for.

EDIT: I'm using peltiers to cool liquid to a set point, I want to ramp them up slowly as the liquid warms up, the farther away from the dew point it gets the more the peltiers turn up. As it gets closer and closer to the set point the peltiers start turning down, which increases their efficiency and should allow me to run lower voltages to keep constant temperatures. When I slam the CPU with a heavy load the peltiers have the capacity to turn up and keep everything cooled, but I won't burn nearly as much energy when idling if I use a control circuit instead of straight PWM.

Plus I'm taking it as a learning opportunity. I went through a crash course in electronics years ago in the military, I got about 2 years of theory crammed down my throat in 3 months. I've forgotten quite a bit of it and we never really covered in depth the formulas or design. I'm trying to correct that and figured this was a relatively simple project to work on and use as a spring board to getting back into learning again!

 

Can you point me to a datasheet or something so I can see how this COP works?

Does this cover it?

 

http://hardforum.com/showthread.php?p=1041236921

Here is a link to a forum where someone has done quite a bit of research and explains it pretty well. He also linked a site that has a good datasheet showing the COP curves at different voltages
EDIT: Scroll to the top post on that page

"Peltier devices work on current, but usually have significant enough resistance so that voltage control is possible.

Peltier devices are one of the few things you do not want to run with pulses, particularly in cooling applications. The cooling effect is proportional to current, but the internal heating due to I2Rlosses is proportional to the square of the current. Starting at 0, increasing current causes increasing cooling. However, at some point the resistive heating due to more current outweighs the additional cooling power of the higher current. More current beyond this actually therefore causes less overall cooling. The maximum cooling current is one of the parameters that should be supplied by the manufacturer.

While maximum cooling occurs at some specified current, efficiency steadily decreases with increasing current. Therefore you don't want to PWM a peltier cooler between 0 and the maximum cooling current. Driving it at the steady current to produce the same overall cooling is more efficient.

Of course the microcontroller regulating the temperature will still produce PWM pulses. These pulses need to be filtered so that the Peltier device sees relatively smooth current. The general rule of thumb is to try to keep the ripple below 10% of nominal, but of course that is just a tradeoff someone picked. Fortunately, this is usually not a difficult requirement to design to."

 

OK. I see how COP works, but it doesn't say anything about 30KHz. There must be a better way to do this.
A mosfet driver is going to slam the mosfet off hard and quickly. Even if you slowed this down to 3 Hz you would still have spikes.

Add ceramic capacitors, quit trying to protect the power supply with inductors, or some combination of those. I just figured out that the big diode is not a zener. D'oh! How about making it a zener and putting a it across the mosfet, or a MOV, or a snubber?

 

I picked 31Khz because it allowed for smooth voltage without crazy sized capacitors (and was easily achievable on the arduino). I think right now I'm sized for about a 2% ripple. I should be able to raise that to around 10% without any adverse side effects, which would be a frequency of around 15khz maybe a bit less? I'd have to go back and look it up again.

At 20 amps I have to hard switch the MOSFET, with soft switching the heat cranks up so fast that it's almost uncontrollable, using a single FET I hit 160F° in about 20 seconds with soft switching and that was at about 12 amps. That was with a 2x3" 1/16 thick heat sink on the FET. I *could* stack the FET's in parallel, but I think that I would need probably around 5 to 6 FET's to keep the heat dissipation reasonable.

Would a pull down resistor on the output of the driver help to slow the gate slam a bit? I know it would slow the rise but I'm not sure if that would really help the issue I'm seeing.

I wish I had a dummies guide for snubber circuits, or knew more about the circuit simulators. Those things are like greek to me! But sure would be invaluable right about now for simulating without spending $100 on test components...

 

A resistor in series between the gate driver and the mosfet will use the capacitance of the mosfet gate to slow down the switching, and it's calculatable. That 2.5 something capacitor on the output of the driver is an unnecessary burden unless the resistor is between the driver and the capacitor, then it becomes a time constant. I'm still barking at the inductors as the cause of the spiking.

 

Smarter people than me will be here tomorrow. You might get a simulation out of one of them...or a whole better approach!

 

So on my list of to do's for tomorrow:
1. Lookup the calculation for the current limiting? Resistor between driver and gate to slow rise and fall time a bit.
2. Read more on snubber circuits and see if I can reduce the spikes enough that way.
3. Possibly add a diode across the FET to protect it while testing since I'm running out of FETs
4. Explore the possibility of lowering PWM freq to help with reducing spikes

I appreciate the help! I'll keep digging into this, any advice / criticism / ideas are welcome. I'll keep an eye on the thread and if I figure anything out I'll post up some results! (Or failures)

 

Yes, the I^2R loss in a Peltier is why you do not want to directly pulse them with a PWM signal.
Thus the need for an inductor in series to smooth the current pulses and a higher PWM frequency allows the use of a smaller inductor for a given current ripple.

 

delta Vgate = DeltaV times e^(-time/RC)
That's the formula for capacitor (gate) voltage change per time based on RC.

 

How about moving your Shottky down, across the mosfet?

The Schottky is fine where it is. It's the free-wheel diode for the inductor current.
But you could add a zener across the drain-source of the MOSFET to clamp any inductive spikes.

Unless you are concerned about EMI you don't need the 0.47μH inductor.

 

Yes, the I^2R loss in a Peltier is why you do not want to directly pulse them with a PWM signal.
Thus the need for an inductor in series to smooth the current pulses and a higher PWM frequency allows the use of a smaller inductor for a given current ripple.

So with an inductor 47uH, would I be able to lower my frequency without going to crazy sized capacitors? I'm scavenging inductors from old PSU's since they are pretty expensive to buy new. I think at 31Khz the circuit called for 9uH or there abouts.