Weber and Coulomb's MX-5

[acmotor]

Pet hate.... rigid bus bars joining cells by their terminals.
I feel that there are many premature cell / battery failures as a result of terminal / post breaking the cell / battery casing.

This is a problem in two situations..
- cell held mechanically in place by its terminals when in an EV application. i.e. cell not well restrained by cell case.
- cell held from expansion, by its terminals to adjacent cell.

Most cells / batteries change dimension (typically expand ?) as they age. They also expand with temperature. If they are held to rigidly by the terminals then case damage (and fluid loss / dryout) can occur. Maybe not today but sooner than you would like.

There must IMHO be some give (1-3mm per terminal) without applying large (edit:mechanical) loads. Use flexible wire or the TS kinked terminals etc.

I think that on Simon's bus bar challenge link, the fair comment was made that something like only 22W loss in 12kW means that you are trying too hard !

[woody]

I think the killer for Ali is its rigidity, a4x4kiwi has kinked (z bend) copper between his terminals which makes good sense - the copper is flexible.

Like the 22W / 12kW, not much point saving 2kg when you have to stop every 500km and replace broken Ali battery links...

[Squiggles]

So, what your saying is that the cells in your battery should be connected with over sized copper busbar that acts as a heat sink. Seems like a simple enough solution.

I know you're only half serious, but this has triggered me to check out something which I've wondered for a while.

Actually more than half serious, if getting heat out of or into cells is an issue the bus bars may be the way to address it.
Lets say you live in Canberra and you need to heat the battery to charge it, you could connect a hair dryer to the battery box vent blow warm air in and the bus bars would transfer heat into the cells. Maybe even set it on a timer to warm them half an hour before driving off into the morning frost.

[Squiggles]

Pet hate.... rigid bus bars joining cells by their terminals.
I feel that there are many premature cell / battery failures as a result of terminal / post breaking the cell / battery casing.

This is a problem in two situations..
- cell held mechanically in place by its terminals when in an EV application. i.e. cell not well restrained by cell case.
- cell held from expansion, by its terminals to adjacent cell.

Most cells / batteries change dimension (typically expand ?) as they age. They also expand with temperature. If they are held to rigidly by the terminals then case damage (and fluid loss / dryout) can occur. Maybe not today but sooner than you would like.

There must IMHO be some give (1-3mm per terminal) without applying large (edit:mechanical) loads. Use flexible wire or the TS kinked terminals etc.

I think that on Simon's bus bar challenge link, the fair comment was made that something like only 22W loss in 12kW means that you are trying too hard !

Simple solution, with quality installation bus bar would be no problem.

[acmotor]

Quite agree !
Quality installation would include smart mechanical design of battery box / restraints AND smart (as TS for instance suggest) bus bars with some flexability.
I hope that was what you meant ...

[Squiggles]

Pretty much, if the battery box and restraining method are good there should be no issue with busbar.

[weber]

Woo hoo! Our batteries are here. 212 x 40 Ah Sky Energy SE cells.



Set vicious guard-cat to guard them while I went to get the trolley:



After helping the driver unload the 8 crates (7 of them a 2-person lift at 54 kg) I'm thinking Zeva might well have left off the winky-face when he wrote "Holy moly what are you and Coulomb doing to that poor little MX5!?" forum_posts.asp?TID=1243&KW=mx-5&PID=13805#13805

More pics, and tests, to follow.

[Johny]

I don't think you guys are keeping up if you think those wooden batteries are going to power a vehicle. BMS? wooden bother.
Actually if you take whatever is in them out and put wheels on them you could make a great little toddler train.

Woo hoo!

I figure you will need a 28kW dummy load to test them at 1C.
I'll think of more useless comments as the day goes on - jealousy does that to me...

[acmotor]

I think the dummy load is more likely to be 50ish !

Hey the cat didn't work !! I checked on google map street view and the boxes are gone !

What was that you said Johny ?

[Johny]

ahh my aching back....

[coulomb]

For more photos of these cells, see this Sky Energy LiFePO4 Bulk Buy this Week thread.

Edit: For voltage sag measurements, see this Thunder Sky impedance post.

[coulomb]

Some metal (angle and a little flat bar) was delivered today.

     

It's difficult to see, but there's actually a fair bit there, over 70 kg I believe. It's 7.5m long. We don't want to cut it till we're really sure of the lengths we want.

Edit: That's actually a downpipe at the left end of the first picture. Why not use it?

Hey, the door opens inwards, ok?   

The shiny stuff near the front is actually some galvanised 30x30x3 mm angle from Bunnings. That was before we discovered the joys of welding galvanised steel.

Edit: With these brackets, we keep the steel out of our way, and out of the worst of the weather. I can reuse the brackets for storing wood or pipe etc. later. That's after we finish the run of MX-5 conversions...    

[coulomb]

We tested the prototype for our BMS today:



The veroboard is downside-up because we wanted to use the veroboard tracks as heatsink for the FET.

We used the actual surface mount FET and opto. The circuit seems to work OK, though the 2.5 V cutoff is a bit gradual for our liking; maybe we need to add some positive feedback to make it snap on and off, like the bypass FET.

The FET got rather hot in linear mode, so that's not right either. It did however switch at very close to the right voltages. The 4.0 V circuit did not work at all. We finally realised that I'd swapped the 4.0 and 2.5 V circuits. Sigh. So a lot of resistor values will have to be recalculated.

Better to discover these things at the prototype stage.

Edit: the LEDs were not bright at all with 1k series resistors. However, with surface mount LEDs, they may be OK, and since these will be on continuously, we don't want the drain to be too high.

[coulomb]

We did some more work on the BMS boards today. Here is the current circuit:



It was decided that the low temperature alarm could be implemented with a single sensor; there is no need for under temperature detection on each cell. That way, we use just the four comparators.

After we drew up this circuit, Weber asked why we need D2; couldn't the opto and green LED be connected to the output of IC1c? Indeed, we'll probably go that way, although it means no red LED, only a green one (which we probably won't populate anyway).

We started testing the new circuit with a combination of veroboard (for the bypass circuit from the FET onwards) and a prototyping board:



This is just the bypass circuit so far. We found that while the bypass turned on at about 3.57 V, it seemed to be oscillating. Weber had his 30 year old CRO with him, but it went fut and sizzle on us. Nothing obvious on opening, so we went to the backup CRO, my 40 year old valve model:

     

Pay no attention to the sideways ghost in the machine    

The trace shows the FET drain. What is going on here? It seems to turn off properly most of the time, but at other times, it seems to turn off before the 3.9 ohm load resistor has had a chance to charge the stray capacitance. This initially happened on a slack power supply of mine, but still happened when connected across the Thunder Sky cell. Here is a clue: the cell voltage looks like this:



Pay no attention to the pointy headed bald ghost in the machine Also note the tiny size of the MOSFET, compared to the 4x7W resistors it will be driving (not at 28 W though, more like 7.5W).

At first we thought the noise was coming from the power supply that we have to nudge the Thunder Sky cell to the voltage we want. But as I was posting this, it occurred to me that the byoass capacitor we were using (the brown one across the chip in the photo) is just a polyester (I happen to have a hundred on hand), and is really just an "audio quality" capacitor.

After I replaced it with a sky cap (actually this one was orange, not the usual sky blue) and paralleled it with a 220uF electro, the LED seemed to snap on and off quite well. However, when I let the cell discharge itself slowly past the 3.6 V point, it seemed to fade out, and sure enough firing up the CRO revealed some high frequency oscillation.

What may be happening is that even a single 3.9R resistor (< 1 A) of bypassing is enough to blow away hundreds of millivolts of that fluffy surface charge that LiFePO4 cells seem to get. So we may have to arrange for a lot more hysteresis than the current 3 mV x (3.6 / 1.235) ~= 8.7 mV of hysteresis.

Actually, the whole idea of "digitally" (slowly) switching the bypass current may be invalid. We may be able to arrange a sort of PWM process that simulates analogue bypassing. But that will be tricky with the cell voltage changing every time the bypass resistor comes on.

Then again, maybe we just need some capacitors across the reference, and/or the positive input of IC1a.

More soon!

Edit: annotated the diagram. Had to blank out part of it to squeeze in under the 100 kB limit.

[Richo]

Try a diode between R15 and R2 (Org - wht wire?).
Then have the bypass caps after the diode.
This should isolate the power section from the low power analogue.
Reduce the noise in the analogue section.
Without the diode the bypass caps are being heavily discharged by the load resistors.
ie not having much effect.

If all else fails try a DSO rather than a CRO

[coulomb]

It's a good idea, except the forward drop of the diode will play havoc with the measurement of the cell voltage. Maybe a small R (10 ohms perhaps) instead.

We'll sort it; just needs time.

[Richo]

Could also move the R3 measurement string before the diode and place a small cap across R4&R5.
Keeps the DC power for analog and reference lower noise.
But measurement section accurate and still responsive.

[coulomb]

Thanks for the ideas, Richo.

Here is the latest circuit:



We keep the R3/R4/R5 divider on the cell side of the new R18 decoupling resistor, to encourage oscillation. This gives us automatic PWMing (Pulse Width Modulation). Capacitor C3 keeps the frequency in the low hundreds of Hz region, so there will be a minimum pulse width of a few hundred microseconds.

We don't have R18 in the circuit yet, but here is the result with C3 in place:



When I connect another 3R9 resistor across the existing one, the PWM ratio changes, but the voltage stayed at 3.57 V. Just like it should.

The nicest part about this is that we didn't design it for PWM operation at all; it did all that itself   

Edit: Well, we didn't design for high frequency PWM; we thought it would switch in the seconds to minutes range. I guess what it's doing now is much the same, except that it switches much faster. The result is the same; the cell hangs around an average of 3.57 V, and the FET doesn't dissipate much power. So it can be surface mount and have no heatsink other than thick PCB tracks.

Also fixed an extraneous junction in the circuit diagram.

[coulomb]

We decided that acmotor's idea of using the "spare" comparator to latch an error signal was so good, we didn't want to "waste" a comparator on the high temperature signal. So we ended up with a one-transistor solution (well, two transistors really, but one is used as just a diode, and is cheaper than a real diode, and real diodes come in the same package as a transistor!).



We spent half a day deciding the value of R8, and figuring out why we were getting a "good" signal when the voltage was low and the thermistor and transistor was hot. It turns out that when we didn't have the diode in Q2's emitter, R8 had to be down as low as 8.2k, and with the thermistor hot, it was loading down the Vref signal too much. The transistor was starting to conduct with Vbe at about 380 mV. With the diode in there, this is doubled to almost 0.8 V, so that R8 needs to be bigger, and the load on the Vref line is reduced. We didn't want to reduce R2, since that would have increased the quiescent current a lot, especially with high cell voltages. We considered going to a series regulator (as opposed to the present shunt regulator), but these are more expensive. The pennies add up quickly when multiplied by 200 or perhaps 220.

The circuit is amazingly complex, though a lot of that is optional LEDs and their resistors. So we likely won't be able to cram it all into our 2 postage stamp available area. We'll know soon.

We tested the over-temperature circuit by soldering the tiny thermistor and transistor to the veroboard, and running flying leads to the prototyping board. It was heated with the venerable hair dryer, testing the temperature with a non-contact IR thermometer. We found it easiest to give the board quick blasts with the dryer, and note the maximum temperature on the thermometer. One set of eyeballs was on a meter or LED, the other on the thermometer. It worked reasonably well; the under-voltage threshold (nominally 2.5 V, but actually 2.47 = 2Vref since you can't readily get resistors that are less than the twenty-fourth root of 10 in ratio) didn't vary much at all with temperature. Without the transistor, that threshold would have increased considerably with temperature, disallowing some perfectly good states like 35°C and 2.6 V on the cell.

Edit: The intention is to arrange T1, Q2, and Q3 to be at about the same temperature as the cell.

[Tritium_James]

Man, it's getting way too complicated. Sorry, it's Friday arvo and I'm bored, here's my version. Very quick & dirty, probably has errors, buyer beware, YMMV, etc, etc. 42 parts in yours -> 27 parts in this one = 40% less time to build it. The MSP430 ($2.38) has internal temperature measurement, and also a 1/2Vcc measurement so you can cross-check your 10-bit ADC & internal reference. Internal R/C oscillator is factory calibrated to 1%, not that you really care. Very low power, and you can probably find a better linear reg with even lower quiescent. Load MOSFET is on a timer channel so it can be PWMed.

[Squiggles]

I'm afraid I am with TJ on this one. I can even long term loan you an MSP-FET430P140 flash emulation tool.

Could probably do more than one cell per board as well!!

[acmotor]

There's more than one way to skin a cat.

Coulomb's will work fine. It also includes O/V protection and reverse voltage protection for the BMS board (D1 F1). This needs to be considered in the uP version as well to cover all forms of assembly and cell fault conditions with limited smoke excape.

Nevilleh's 8 cell uP monitor in the BMS thread shows promise too.

TJs uP sketch is fine.

Actually, I'd be happy with cells that just work and don't bother me unless they have a problem 'cause there is not much I can do about them if they go flat !

[coulomb]

It seems that the circuit of a few posts ago has at least two problems with the error latch. We neglected the voltage drop across the comparator outputs when driving a LED. The effect is that the latch resets at too low a voltage, around the bypass voltage. The fix is a simple resistor change, except that TJ shamed us      into using resistor arrays, so now we have constraints on what resistors can have what value. It looks like we can just swap some resistors.

It's a pity that we'd just finished laying out the board! (With many tidy-ups still to be done):



The second problem is more serious. When the bypass current comes on, the change in comparator voltage (despite bypass caps everywhere) seems to trigger the latch to reset itself, at least on our breadboard. We'll sort it.

[acmotor]

Wouldn't reaching bypass be a logical point to have the latch reset anyway ?
The main function of the latch would be to record U.V. (when EV driving under load) and there needs to be some way to reset it ? (without a screwdriver)

[coulomb]

Wouldn't reaching bypass be a logical point to have the latch reset anyway ?

We assume that regen braking would likely go to the bypass voltage and even to controller cutback a lot of the time. We don't want pressing the brakes in the driveway right at the end of a long trip resetting the important information from the trip.

However now that I think about it, if a cell is undervoltaging (the latch also sets on overtemperature), the pack is likely at a rather low SOC, so bypassing would be unlikely with regen. So maybe it doesn't need fixing   

Our idea was to do a brief overcharge of the pack (for say a second) to about 3.75 VPC to reset the latches, at the end of a garage charge. The pack would be rather high internal resistance at that point, so it shouldn't take much energy to do that. It would be nice to get the original concept working.