Linear Tech 12-cell battery monitor/BMS

[CamLight]

Very, VERY nice methods! Congrats.

[mtx]

Hi, seeing that like me you guys have been working an BMS's using the LTC's I was wondering if any of you have had problems (Like me...) with devices being unrealiable. For example I'll have a perfectly working board and out of nowhere one day I'll turn it on and find that the LTC is overheating or the ADC reading are wayoff on the entire chip on my master or slave board. I'm currently using the -1 version with one master board and one slave board connected (total 24 batteries) with board-to-board connectors, my micro is optically isolated from the LTC,I'm using external fets and have followed all possible HW recommandations from the datasheet and LTC Eng.

For now my best guess seems that maybe the damage is being done when I connect my batteries to the board though the connector, in the past I was blowing LTC's this way and fixed the problem by connecting the last battery of every LTC through a switch in order from master out only after all batteries have been connected. I'm thinking of automating this using some fets in the future.

Any suggestion is welcome, thanks.

[methods]

I suspect that you will find that it boils down to silly things like loose solder connections etc. I have probably thrown away 30 LTC chips during my testing and evaluation stage but once I got the process down all of my troubles went away. Keep the board sparkling clean - be very weary of trace spacing, especially between the last few cells (10, 11, 12) and ground. If you implemented the RC circuits each one of those caps runs to ground so..... check your clearances.

99% of my troubles have had to do with "silly stuff" - check, check again, and if that does not work wipe the part of the board, clean up the pads, and start over.

A hot LTC means stray currents - this can happen easily with the -1. Are you using this in a vehicle? I found the -1 to be unacceptable for noisy environments.... as soon as the noise kicks up the stray currents screw up the readings. It can be solved......... but it is a hell of a lot easier to just switch to the -2 (the arguments for which I presented above).

-methods

[hardym]

For example I'll have a perfectly working board and out of nowhere one day I'll turn it on and find that the LTC is overheating or the ADC reading are wayoff on the entire chip on my master or slave board. I'm currently using the -1 version with one master board and one slave board connected (total 24 batteries) with board-to-board connectors, my micro is optically isolated from the LTC,I'm using external fets and have followed all possible HW recommandations from the datasheet and LTC Eng.

I had similar problems following the LTC sample ckt. I had first thought that it was the order batteries were connected to the board. I could crash a LTC chip in minutes. LTC's sample ckt did not take into account is stray inductance on an external load resistor, such that when switched off would generate a flyback thru the LTC pins. There is an 8V zener protection inside the LTC which will short if too much current passes, causing erronious voltage readings and heat. Strongly suggest a simple diode (IN4935) in parallel with your external load resistor. At .04 each, it's cheap insurance. This recommendation is based on a review of my ckt by LTC engineers. After adding the diode, the boards were very stable.

Mark.



I

[methods]

You are saying that you place an *additional* flyback diode in parallel with the 6.1V protection zener prescribed in the reference design?
(EDIT: I meant to say in parallel with the resistor and in addition to the 6.1V zener across the pins)

So are you saying that these additional diodes will turn on sooner thereby protecting the internal diodes?

The reference design clearly calls out 6.1V external diodes -> D16 - D27 and this is like the 3rd iteration of the design...
Reference Design

I think that careful selection of those external zener's meets both needs: Clamps the input to 6.1V (WAY below the internal zener) and they can also be selected so as to conduct sooner as well (thereby serving to short any flyback). They are 500mW even in the tiniest packages.

Were you running external 6.1V zener's when you started blowing out channels and decide to add flybacks? This just does not add up for me because I am not seeing the same problems with my wire wound resistors. I would argue that it definitely does cost you something... 0.04 is just the cost of the part. Multiply that times the cost of board population and the cost of additional board space and it starts to get expensive.

-methods

[methods]

Illustration for clarity

-methods

[hardym]

Here is a clip with the diode
Latest reference design:

refdesignclip.JPG

(22.11 KiB) Downloaded 422 times

Early Reference Design:

refdesign-early.JPG

(12.49 KiB) Downloaded 404 times

I'd like to review and extend my previous comments.

The first diagram is what methods posted as the latest reference design. This looks like an ok desing and probably does not need the diode across the load resistor as shown.

The second diagram is the reference design from LTC that i was given a year ago, with the 8v protection zener on the WRONG side, which provided NO protection from the flyback to the LTC chip. No, I am not high.

A flyback protection diode is necesary the original (bottom) ckt, but may not be necessary with the ckt on the top. YMMV. My original statement is still true: without the protection diode I could fry a board in minutes of testing. with the protection diode, the boards were pretty stable.


Unsolicited, unrelated comments: Any installation that is mis-wired will likely blow the LTC. With this type of architecture, there's not much to prevent it. So I dont think this LTC6802 is a good architecture for a distributed battery system (like in a large EV). May be better suited for 12 cell embedded modules, like a UPS or off-grid power storage. Oh, and I was extremely unsuccessful using the LTC-LTC comms solution in the LTC-6802-1. I used separate PIC controllers for each LTC and opto-isolators. Not elegant, but it worked reliably.
Mark.

[mtx]

Sorry it took me so long to get back to you all but I did not receive any email notification that there was an update to this post... Strange I'm sure I selected the notify check?? Anyways....

Thank you all for your replies, over time I have made two versions of the board, the first version is using the "refdesignclip.JPG" without the diode in parallel to the resistor and the second version (and current) is using "refdesign-early.JPG". I think that "maybe" I have resolved the issue in the last few days by putting a softstart mechanism that closes the connection to the supply of the ltc's in sequence and lastly closes connection to my isolated flyback circuit that powers the mcu. I'll update on this matter after I finish some more testing and am sure that all is really working properly.

Unfortunately I am currently using the -1 chip and yes I too believe that if I did it all with the -2 I would have probably resolved a lot of the headaches faster but when I started on this venture the -2 was not available in my country. As for the communications I too had problems initially with stability and signal noise but I must say that fine tuning the comm. routines I have been able to keep packet loss down to a minimum (0.1% packet loss over a sample period of 100000 readings) while both running the vehicle and charging, I am confident that this is not an issue for my application.

I am currently using this system to manage electric scooters because like was pointed out this is absolutely no good for larger applications where the number of batteries & currents get too high. Actually in my opinion for very large battery packs non of the commercially available solutions are very well suited, for these types of applications I have developed a custom solution that works very well the only reason I don't use it for managing these smaller battery packs is because the cost is higher and cannot be justified for the application.

[EVan]

Unsolicited, unrelated comments: Any installation that is mis-wired will likely blow the LTC. With this type of architecture, there's not much to prevent it. So I dont think this LTC6802 is a good architecture for a distributed battery system (like in a large EV). May be better suited for 12 cell embedded modules, like a UPS or off-grid power storage. Oh, and I was extremely unsuccessful using the LTC-LTC comms solution in the LTC-6802-1. I used separate PIC controllers for each LTC and opto-isolators. Not elegant, but it worked reliably.
Mark.

I've used 4 of the LTCs with a distributed battery system in my full size EV (23kWh battery) and it's working well so far (7000 miles since the start of the year). In fact the BMS hasn't been touched or reset since April.

I am using the built-in current mode SPI comms between devices. It works. BUT, the devices are central in one unit. If the data path is longer than a few inches in a high power application then I agree it doesn't have much chance.

[mtx]

On my first board I was using board to board connectors for the comm. between LTC's so the overall distance was minimal instead on my newer design I am using a 10 pin flat cable about 15-20cm long and I have no problems like I mentioned in my previous post.

[methods]

Patrick, how about using silicone caulk from a caulk gun. That should have the pressure needed to flow down the channel.

Thanks - you were right. In the end I selected a non-acidic cure COTS caulk fit for a gun. Cost 1/10th the price and worked killer.

Beautiful and it works great as well! For the silicone you might find a bead on the outside of each leg works as well as a much larger amount underneath the resistors. The resistor bond would then be in shear. Adhesion to the board is often not a problem.

Thanks to you too sir - After some testing I found this to be a much more effective method then filling a big blob under the resistors. In the end I ran "rails" down the PCB for each row of resistors. The legs of the resistors landed in the "rails" and squished out a little. I then ran a Q-tip down the sides to smooth the bead. I also ran a bead along the top to tie them all together like one large mass and then baked them out in the oven. Came out great - not a single problem yet.

Now... This is not to say that there have not been problems

I have 3 units sitting here on my desk that are missing (or at some time have lost) their magic smoke. To be clear - none of these were "mysteries" - all had a clear cause and effect - sometimes a long time in the finding - the cause being either me or the customer in all cases.


1) My prototype - burst into flames at one point when I ran it at 150V 20A without letting the PCB cleaner evaporate completely. I have since rebuilt this (several times) and it is still in service

001_Methods_BMS_V3.0_Smoke.jpg

(103.65 KiB) Downloaded 279 times

002_Methods_BMS_V3.0_Smoke.jpg (69.94 KiB) Viewed 692 times

003_Methods_BMS_V3.0_Smoke.jpg (103.73 KiB) Viewed 692 times

2) Not to be outdone - the customer sent me the pictures below. After a lot of head scratching the root cause of this failure was finally discovered... and it was a real good one! So very sneaky... The Charge leads and the Pack leads were inadvertently swapped. This means that instead of my IRFS4115 fets acting as PWM switches, their body diodes acted like heaters! The body diodes on these fets develop 1.2A and the customer was charging at 25A !!! That is 30W in a 1cm square area Needless to say everything was baked out including the (ever so important) conformal coating. The BMS actually completed its charge and then later - after a valiant effort - vomited its charge wires and died. Actually - it really is not bad at all. The BMS is still fine, only the charge fets and ceramic caps have to be changed out.

001_Customer.jpg (18.74 KiB) Viewed 692 times

002_Customer.jpg (29.79 KiB) Viewed 709 times

003_Customer.jpg (24.06 KiB) Viewed 692 times

3) The third was the inevitable "miss-wire" where (it at first appeared) that the balance taps were plugged in out of order. It later was found that they actually were *not* plugged in out of order - but actually a freak accident occurred where the balance taps got crushed into a parallel adapter board which shorted out a bank of 5 cells. Long story short - a few channels got damaged but the BMS still works and needs only minor repair.

But... That is just the fun stuff

I have learned a TON along the way and I have quite a few tips and tricks. I am keeping the details of those to myself - but in the spirit of open source, here are a few VERY IMPORTANT tips that I will share with you.

1) Manually shut down all shunts before you take your A/D reading. Just shut them all off, poll the A/D's, then turn them back to whatever state they were in. Yea, yea ... I know the LTC will do this for you - but if you do some testing you will find that you can greatly improve your S/N by doing this.

2) For the LTC6802-1 you must do the above - here is why: The built in function to turn off the shunts is not aware of how many LTC chips you have in series. This means that when you try to read cell 13, the second LTC has no way to tell the fist LTC to turn off the shunt on cell 12. You either get what I am saying right now or you dont. If you dont - hit the books. If you do, then you understand how important it is to address this issue. Any time an A/D sample is taken you must turn off that shunt + the shunts above and below your sample

3) Turn off any PWM that you are running when you take A/D readings. TRUST ME... Your S/N will go WAY up. I have an algo that basically monitors the state of the charge PWM and turns off the PWM when a critical measurement is taken. When PWM is at 100% - obviously there is no need. On the other hand - once you are balancing.... You may find that you go into oscillation with your shunts if you are not getting really accurate readings. To give you an idea - I now balance to 2mV. Yea - thats right -> 2mV on a 25S 80Ah pack.

4) Really, really, pay attention to the noise you are creating with any PWM circuits. Switching 100V @ 25A is no joke -> That is 2.5Kw, thru many feet of inductive cable, and from a capacitive source. This can reap insane havoc on your switching system. In the end I solved the problem by throwing >10uF of 250V ceramic capacitance at it. This circuit was an electrolitic destoryer!

5) Design your prototype for rework.... use chip sockets for any dip parts, dont bother populating unimportant circuits, dont conformal coat, etc.

6) Use an acrylic conformal coating if you plan to re-work. Use a silicone conformal coating once you are confident.

7) Keep tight control on your firmware versions! Now that I have 3 versions of the BMS out there and a 4th in the works it is REALLY IMPORTANT that I send out the correct firmware updates. Really - the three versions are not compatible in any way - especially since I swapped to the -1 between versions 2 and 3. I have a TON of different versions out there for different applications too - some for dyno testing, some for troubleshooting, some for assembly checkout, some for my car specifically, some for...... Keep them straight and keep them safe!

Ok - that is the extent of the bad.
The BMS works beyond my wildest expectations and I could not be more pleased. Sure I had to do some hacks - but those are easily fixed with a PCB spin. All and all - things are working out splendidly!

Below I will post some fun pictures.

-methods

[methods]

003_Methods_BMS_V3.0.jpg (85.77 KiB) Viewed 710 times

Some early hacks to get the PWM working. Turned out I went back to 4 fets and about 6uF of ceramic capacitance

011_Methods_BMS_V3.0.jpg (114.17 KiB) Viewed 710 times

028_Methods_BMS_V3.0.jpg

(119.51 KiB) Not downloaded yet

029_Methods_BMS_V3.0.jpg

(119.45 KiB) Downloaded 280 times

031_Methods_BMS_V3.0.jpg (71.53 KiB) Viewed 710 times

mmmmm.... The smell of BMS baking in the oven!

037_Methods_BMS_V3.0.jpg (93.54 KiB) Viewed 693 times

043_Methods_BMS_V3.0.jpg (80.91 KiB) Viewed 693 times

044_Methods_BMS_V3.0.jpg (58.4 KiB) Viewed 710 times

061_Methods_BMS_V3.0.jpg (39.7 KiB) Viewed 710 times

Who is hungry?

055_Methods_BMS_V3.0.jpg (72.29 KiB) Viewed 693 times

-methods

[kfong]

Excellent work Patrick!

[Hyena]

Nice!
Did I miss somewhere that you're selling these ? Or are they beta tester units ?

[methods]

I am not selling these to the public.

If I ever get the money to do an ebike version it would look very different. I would do 12S modules that daisy-chain in a much smaller housing with a very small fan. I would probably eliminate the entire USB interface (to save power) and just have the system work stand-alone.

There are a lot of aspects to this design that make it undesirable for ebikes - not the least of which is its huge size and high price tag.

I am actually going *bigger* with the design for the next version. We are going to integrate more features into the housing and increase the charge handling to 150V 50A. That is what my batch of IRFB 4115's is for that I am sharing in the For Sale section.

-methods

[fechter]

I have 3 units sitting here on my desk that are missing (or at some time have lost) their magic smoke. To be clear - none of these were "mysteries" - all had a clear cause and effect - sometimes a long time in the finding - the cause being either me or the customer in all cases.

Those look like pretty impressive burns. That's how you learn

Yes, that PWM stuff really raises hell with all sorts of things. At least you have the option to turn off things during the measurement. Trying to maintain accuracy to within a few mV with kilowatts of noise flying around is difficult.

Anyway, looks like you're making progress. Very nice

[methods]

I have been paying for my education alright

I have a lot of experience with firmware and electronics but I am definitely still learning about power engineering. Lately I have been trying to wrap my head around heat-sinking requirements. Luke has been helping me to try and estimate how many fets and how much heat sink I need to do a completely passive system. When you add the 50A 4khz PWM to the mix the power handling requirements go thru the roof! I think we have a handle on it and I am going to start building prototypes soon. Unfortunately with V3.0 I never had a lot of time to prototype so I ended up having to modify the circuit after the PCB's were spun out.

Having the option to turn off the Shunts and the PWM makes all the difference in the world at the last stage of balancing. Once all the cells are around 4.18V (as you know) just about any balance circuit will go into oscillation if there is any kind of noise. I chased my tail for a few weeks until I figured that out.... My shunt display used to play like a ghost piano.

I was looking at your latest creation..... Looks like you are using a similar zener approach to protect your TC-54 chips. I was not aware that you were popping those but it makes sense. I know you are the cheapest bastard on the planet so those zeners must really be necessary or you would not have them.....

-methods

[methods]

Ouch.... Built an 18S 10Ah pack for a ride with a new Space Monkey and something went wrong... I think I might have actually reversed cells for the first time. After charging for 30 minutes the lowest cell was at 1.5V. Now after about an hour I am still not even up to LVC

The two packs that were low were HOT HOT HOT

001_FUBAR.jpg (33.05 KiB) Viewed 786 times

Ignore the cells above 18
19 and 20 are short circuits, 20 - 25 are open circuit

-methods

EDIT: By "Went wrong" I mean that I did a crappy job building the pack. Nothing happened with the BMS - there was no BMS - As it obvious from my results

[liveforphysics]

Wait... are you saying the goal isn't to run your LiPo cells down low enough to reverse cell voltage?

[methods]

It is official - for the first time I have ruined lipo cells out of pure nOoBieNess
I think I might actually put a BMS on my "loaner" bike now

-methods

[methods]

I have stopped reading PM's from people asking for help on the LTC.

On a brighter note - I have been patching up those units that were damaged in the field. The protection circuitry worked great - on one unit I only needed to replace a 100 ohm resistor and on the other I need only to replace the PWM circuits. Isolating each section of my board was a wise move in hind sight..... Did not lose a single LTC in all that carnage

The vertical mount shunts are still good too - no failures yet.

My only real complaint is that I tried to stuff everything into such a small box... Makes rework a living hell

-methods

[heizal]

hi all.
I'm heizal from Malaysia.
actually I'm beginner in battery management system currently do project using LTC.
I would like to ask a question.
>May I know what is the ideal voltage different between each cell so that can start doing discharge for balancing all the cell?

tq

[amberwolf]

May I know what is the ideal voltage different between each cell so that can start doing discharge for balancing all the cell?

0V

[heizal]

I hope someone can help me.
1) what is allowable tolerance of voltage between each cells so that it can be balance? the voltage variation rapidly change and if 0V difference to be balance the cell, the discharge operation will do continously.
2) when stack more than 1 board, I found that the voltage reading in second board is different. Voltage reading is lower than actual. Is it my firmware design wrong?

Hope someone can help me as i'm beginner in this field.
TQ in advance

[methods]

I hope someone can help me.
1) what is allowable tolerance of voltage between each cells so that it can be balance? the voltage variation rapidly change and if 0V difference to be balance the cell, the discharge operation will do continously.

Damping

0V is fine - but your circuit will just be under damped or in oscillation
Anything greater is also fine - it will just drive your accuracy - right?

How about 20mV?

I do 3mV.... But then I also do a lot of things to make sure I take good readings - things like turning everything off while my A/D's are settling.... Taking the naive approach will result in very noisy readings (i.e. allowing the LTC to control the shunts while you take readings). I turn them all off manually -> Resulted in significantly less oscillation.

2) when stack more than 1 board, I found that the voltage reading in second board is different. Voltage reading is lower than actual. Is it my firmware design wrong?

Hope someone can help me as i'm beginner in this field.
TQ in advance

You need to include a lot more information here.... but right off I would guess you have a grounding issue between the boards. Could be firmware, could be hardware, could be wiring. My guess is that you are rigging together demo boards with too long of a ground lead. Is is just cell number 13 that reads low or 13 - 24? If it is 13 - 24 it would probably be a firmware issue right?

Think it thru and include all of the details. You will know that you included enough detail when you answer your own question before finishing the post

-methods



P.S. From that WIKI -> this is what you want.

Critical damping (ζ = 1)

When ζ = 1, there is a double root γ (defined above), which is real. The system is said to be critically damped. A critically damped system converges to zero faster than any other, and without oscillating. An example of critical damping is the door closer seen on many hinged doors in public buildings. The recoil mechanisms in most guns are also critically damped so that they return to their original position, after the recoil due to firing, in the least possible time.

In this case, with only one root γ, there is in addition to the solution x(t) = eγt a solution x(t) = teγt:[2]

x(t) = (A+Bt)\,e^{-\omega_0 t} \,

where A and B are determined by the initial conditions of the system (usually the initial position and velocity of the mass):

A = x(0) \,

B = \dot{x}(0)+\omega_0x(0) \,