"Zephyr" - Finally, the "v4" Fechter/Goodrum/Hecker BMS...

[cor]

The elegance of the Zephyr BMS is a very simple circuit for each cell, while still protecting each individual cell for over- and under-voltage; allowing full protection of each cell by shutting off charging as well as discharging (throttle) before any cell reaches its danger zone - at least in theory.
There are some limitations however:
1 - If the BMS electronics or wiring fails, then this may not be detected, so the affected cell(s) are not longer protected
2 - if a cell suddenly goes to zero volts instead of gradually dropping below 3 (or 2) volts, then there is no protection and this also leads to other problems.
3 - if a pack is not charged soon after hitting the low voltage threshold, then the TC54N continues to indicate LVC, running the cells down quickly.

The problem with wiring can happen with the inter-connection between the 6s/8s/12s boards that sends the combined LVC and HVC signals from one board to the next, if a wire breaks then all the cells connected to the boards before that point will lose protection, (whether over-charge or) over-discharge will no longer be communicated for those cells. If a wire breaks between LVC output of the last board and the throttle (or brake input on the controller) then this means that the user is no longer notified of a cell reaching the lower limit by cutting drive power. If the BMS is visible to the user, then LEDs on the board may signal to the user that something is amiss, but it is not likely that a user will see a warning LED on the BMS while riding. One possibility is to connect a warning LED closer to the field of visibility, such as on the handlebars (most eBikes already have a set of indicator LEDs there) but the problem is still that even the wires connecting these LEDs can break. Also, even if the LED lights up, that does not protect the cells unless the user decides to reduce throttle.
It is possible to make a fail-safe system, but the complexity will increase much and it is not easy to test each circuit, most likely a microprocessor will need to be added to cycle through all channels and verify that the LVC signal is enabled on each channel detecting low voltage (for example by adding a resistor between the cell and the input of the TC54N plus a resistor and transistor to ground, to simulate a low cell voltage). The throttle (or brake) signal should be clamped to ground until all circuits are verified to work. Commercial BMS'es might implement this level of self-test, but it seems over the top for a DIY system.
One thing that may be done simple and easy is to add a connector to the open LVC input on the last board in the string, so that occasionally a plug can be inserted to short the signal there, then verify that the controller cannot engage due to the trottle (or brake) signal being clamped, verifying that all intermediate LVC wiring is OK.
Another even more sneaky type of wiring failure is when one of the wires between pack and BMS fails. If one of the "end" wires fails then we lose the corresponding channel, so that cell is no longer protected. If one of the intermediate cell connections fails, then the average voltage of the two cells may be OK while the actual cell voltage is too low (or too high) and the two channels might also fail to properly indicate even when the average voltage reaches LVC, because due to the votlage dividers on the 431 references, the voltages will divide perfectly between the two channels until one channel hits LVC which will then push the other channel to the HVC condition (4.2V) while the firs channel voltage suddenly drops from 3V to 1.8V. This may be too low to reliably indicate LVC, so the two cells that share a broken wire to the BMS will then be unprotected.
Then there are cases where the electronics fails, either because a component fails or something affects the BMS board (moisture, chip of metal, loose wire) and causes the fuse to blow. This will go undetected as this channel will simply stop working, so that cell is no longer protected. Also a bad solder joint or component that fails open due to thermal, mechanical or electrical shock will render the BMS inoperable without warning.

The situation that a cell suddenly goes dead will lead to two problems: since each channel of the BMS is powered by the cell that is being protected, this means that protection fails when a cell fails. The other problem is that the total voltage of the pack drops, which can lead to problems while charging, either over-heating the BMS shunt resistors or the Charge Control FET. In extreme cases the dropped pack voltage may cause the Charge Control FET to exceed its voltage rating, since it is designed to only carry the difference between charger voltage and pack voltage, but this is not likely with a 100V FET rating. It is more likely that the increased voltage difference will cause overheating once the first channel requests throttling back the charger. The danger here is that a failure will typically lead to a short circuit in the FET (typical failure mode of most semiconductors). This means a loss of control over the charger and with one (shorted) cell less in the pack, the remaining cells will get more and will be severely over-charged...

When the pack is run down to the point where the BMS disables the discharge by enabling the LVC, the low cell will continuously see the drain of the opto-coupler LED to enable LVC. Even though the 1k Ohm resistor will limit that drain to less than 2mA, the cell will see a drain that will discharge it down to about 1.2V (the opto's LED threshold) in a matter of days. So if you forget to put the pack on charge for only a few days after running it down as far as the BMS will allow you to go, you might permanently have lost the pack. It could be wise to add a disconnect when the drain should stop, or at least add a diode in line with the LVC opto to stop draining the cell at about 2V so it has a better chance of survival, even though this change will actually cause a reduction of the reliability of the LVC, because if an already well drained pack suddenly needs to dish out a large current and the cell sags instantly from 3+ to below 2V, there will be no LVC signal because the voltage is already too low....

There probably are a couple more failure modes for low cell voltage that I overlooked, but these are what popped in my head when going through the schematic of the channel and system. In case of questions/corrections: fire away!

[cor]

Just like the undervoltage case, the over-voltage can happen with the same type of causes:
1 - electronics or wiring problem
2 - cell failure: open circuit
3 - charger failure

If the over-voltage (HVC) signals are not reaching the charger controller, then cell(s) will be over-charged without the user knowing, this may lead to fire or at least to quick deterioration of cell performance. In particular if the "Any" HVC signal is not throttling the charger back, then some cell(s) will over-charge.
Also if a wire between pack and BMS fails, it will not detect an overvoltage until the average voltage on the two adjacent cells reaches max and if it happens to be one of the "end" wires then the channel will be dead and no protection will be present.
If electronics fails, then the HVC signal may no longer be generated or, if the failure is at the charger controller, the arriving HVC signal may not lead to throttling back.

Open cell failure is a nasty type of fault, during load the cell will try to go negative (reverse) and this might completely blow the associated channel electronics, for example the TC54N. During charging the HVC signal should throttle the charger back so the result will be that the pack will not charge at all.
During regenerative braking, the throttling is often not present, so the voltage will increase uncontrolled until the zener protection diode will shunt all current. If this situation lasts for more than a very brief moment, the zener will overheat and fail shorted, which will blow the fuse. After this there will be no more indication that this BMS channel failed unless the fuse (and electronics) are checked when replacing the failing cell or pack.

The situation of a charge controller failure is already discussed above, there is still the case where the charger starts delivering higher than usual voltage due to a failure in the charger or connection to a wrong charger. It is possible to warn for such failures (as well as for excessive low pack voltage that may indicate loss of a cell) by measuring the voltage drop across the charge throttling FET. Too high voltage drop will overheat this FET and may result in loss of control, causing over-charging of the pack at may charge current indefinitely...

[heathyoung]

On regen, unless you have completly full cells or really high resistance cells, the chances of exceeding the the voltage is very unlikely - especially if you have configured your controller correctly.

[cor]

For good measure, I should say that even after writing the previous 3 posts about issues that can develop with this BMS, I am still planning to build one of these for my eBike, because I like the elegance of a minimalistic BMS and I will likely, happily, run it on one or more of my packs. I am just saying that you won't find this BMS on a commercial EV because there will be a much higher requirement for fault detection.
Still - Richard, Gary and others: cheers and thank you!

[cor]

On regen, unless you have completly full cells or really high resistance cells, the chances of exceeding the the voltage is very unlikely - especially if you have configured your controller correctly.

It is an issue that I hear regularly from EV'ers that live on a hill - either they must take care to never charge their pack to full, or they will need some way to reduce/defeat regen when they drive/ride away from home, down the hill.
I have owned and driven an EV with AC motor and controller for about 3 years. It was capable of sending 200+ Amps back into the pack, which meant that on the lead-acid batteries that I had, the regen had to be disabled until SoC dropped to about 85% or the regen would easily push the pack voltage above the emergency cutoff voltage, which would mean a shutdown of the controller until power-cycled. Not a fun way to merge onto a Freeway (this was a S10 truck) and I was living in flat Silicon Valley! After 3 different events where the controller shutdown due to unexpected hard braking in the first few miles, I had finally overcome my wish to squeeze all juice back into the pack and reconfigured the controller to disable regen until SoC was low enough to allow the pack to soak it in.

[cor]

In continuation on the BMS issues list, one thing that is almost universally a fault detector is an overtemp sensor.
Whether it is a pack getting too hot (See the measurements taken of the "fake" Leadershobby LiPoly repackaged Tekocell packs) or the charger overheating; the charge-controller getting too much voltage and current to regulate or a large bank of shunt resistors shunting too long or in too high ambient. Even a motor controller getting too hot might be a reason to trigger a BMS event and reduce throttle, though it is actually outside the BMS context.

[cor]

BTW, what is the reason to put an AP431 in the BMS instead of the usual TL431? Was that a price or availability issue or is there a technical reason to go for the alternative manufacturer and type? I guess this is a question for Richard.

[fechter]

The TL431 has a slightly lower standby drain, but more importantly we found it has a slightly different turn on point that allows us to get closer to the desired set point with the available resistor networks.

One approach to signaling I looked at before is to use cascaded transistor switches instead of opto couplers. In this arrangement, if any cell triggers it trips the cell below and so forth to the end of the chain where it activates the HVC or LVC. This does a few things; the cell drain during activation can be much lower, a few uA, and there is a fail safe for any cell that becomes disconnected as this will trigger the chain. There is also no real limit to the cell count and there are fewer issues with creepage distance on large packs as the voltage differential from cell to cell is low. The Seiko chips use this kind of setup. Downside is way more parts per cell , possibly more susceptible to noise or contamination on the board and some propagation delay in the chain.

The Zephyr, being a centralized system, has little chance of a disconnection in the outputs of the optocouplers as they are all on the same board or in the same box in larger setups. Single failures of the tap wires (other than the end cells) will result in a pair of cells acting like one cell, but this still gives a fairly good level of protection.

All of the most recent Zephyr control boards have a temperature sensor that will suspend charge if the control board overheats. There is no provision for cell temperature monitoring, but that would be a desirable feature if it didn't take too many parts to implement.

[cor]

Hi Richard,
You chose the AP431 but you mention the advantages of the TL431, so I am confused.
I have been looking at combining multiple outputs to save on optos.
For a maximum of 8s you will have a max 8 x 4.2 = 33.6V which means that usual transistors and the TL/AP431 will allow combining 8 channels into single optos, which means that instead of 8x3=24 optos only 3 optos are needed for each 8 series cells.

I also like to make the LVC signal active, meaning that the throttle clamp or brake is held active unless the wiring is intact and the battery voltage high enough (above LVC), so that a broken LVC wire does not defeat the low voltage protection and you damage your expensive pack.
What this needs is an extra transistor to invert the clamp/brake signal at the throttle signal and turn the default active signal back into a signal that only clamps when there is a problem. Also a switched power input is needed, so that the BMS does not continue to draw power when the bike is switched off. This can actually reduce the power draw of the BMS to zero.
The last thing I like to do is shrink the whole BMS and Charge controller by making it SMD technology. There are plenty PCB houses that will supply a relative small number of (populated) PCBs for a reasonable price. I have dealt with PCBfabexpress in the past, I often pass by them on one of my commute routes.
I have a few sketches and drafts of how to add this to your BMS, let me know if you like me to share them.

[fechter]

All the '431 variants are pretty close in performance but as I mentioned, they have slightly different trigger points.

To combine multiple 431's into a single opto I think you'd at least need some diodes or transistors to prevent applying negative voltage to the output. You'd also probably need different resistor values for the series resistor feeding the opto. On the LVC side, the TC54 is rated for a maximum of 12v, so again you'd need level shifting transistors to go more than 3 cells. By the time you do all that, the cascaded transistor chain will have fewer parts.

The Chinese BMS just use a single transistor on the ouput of each cell circuit that pulls up a common line on the control circuit. These are coupled by 10M resistors to keep the drain low.
This approach works and is sort of a comprimise between the extremes.

I agree a fail-safe cutout is a good feature. The latest Zephyr circuit has an option for a mechanical relay output with both normally closed and normally open contacts. While the relay is power hungry (about 10ma) it makes it very easy to tie in to most any motor controller. The NC contacts can be used to complete the throttle power circuit or controller power circuit so that a fault in the wiring causes the power to stay off. Yes, you could also do this with a couple of transistors to save power.

The BMS really only needs to be active when the controller is on (discharge) or when charging. The Zephyr senses when the charger is active and automatically turns on the power hungry parts of the control circuit. The discharge side of things is always on.

Something more like the CellLog would probably be a better direction to go. The CellLog just doesn't have a sleep mode that works and needs a provision for ganging more than one in series. Other than that, it seems to work well and a single MCU monitors 8 cells and can be programmed by the user. In theory, there should be a way to make the processor sleep most of the time during standby and take a measurment only once in a while. When the voltages start changing rapidly during charge or discharge, the MCU could stay awake.

[cor]

Hi Richard,
I was thinking more of using a key switch signal that also activates the motor controller, as signal to the BMS to make the LVC circuit active and the controller can deliver power to the motor as long as the "notLVC" signal is active. If a wire breaks or LVC is detected, then the controller will cut out. Some controllers have an automatic cruise control, so depending on the controller it also needs to affect the brake input (which will defeat the CC)

For the overvoltage signaling, to generate an "AnyHVC" signal it is only needed to connect a set of resistors to the TL431 (indeed different values for each cell and only up to 8 series to stay below the max 36V limit on the TL431. The "AllHVC" signal to stop the charging will require one transistor and for each cell two diodes to separate the cell circuits. For protection of the HC54 I could not find a way to do it with less than 2 transistors: one transistor plus a diode on each cell circuit and one transistor to drive the opto until one of the cells signals a LVC.
I am less concerned about the charge controller failing to stop charging, because I expect that BMS will typically get integrated with the charge controller and the charger should already have a max voltage to stop charging when all cells are full, so as long as the BMS keeps the cells balanced, the charger control should just be a secondary protection.

If one of the cells fails shorted, its LVC control signal will avoid the motor controller from drawing power so it is likely to be detected immediately and charging should not be attempted in such a case. I did not add a protection against charging if a cell has failed shorted, other than a test input for overvoltage detection test, where the resistor divider is bypassed to make the HVC circuit detect an overvoltage as long as the cell is above 3V. That test can detect a failed cell. When the BMS and Charge Controller PCB is integrated and done in SMT then the price of a transistor or diode becomes almost neglicible, but the optos due to their physical side and creep distance requirements will still be costly parts, that was one of the main reasons to add electronics to reduce the nr of optos. Also the automation of SMT production reduces the labor involved with building the boards, so as soon as nrs increase that is a requirement to keep supplies meeting demand.

[fechter]

Here's an example of a transistor chain arrangement, courtesy of Randomly. If any cell triggers, it 'breaks' the chain and triggers the output. If a connection comes loose or if a cell dies, it also breaks the chain. This gives a good fail-safe for connections. This also eliminates optocouplers all together. No problems with creepage distance if properly laid out.

I came up with a similar arrangement (but can't seem to find the file) that would have slightly lower standby drain and no real limitation on the cell count.

Randomly LVC3.jpg (105.63 KiB) Viewed 659 times

[cor]

Replace Q8 with an opto and you can have "unlimited" paralleled blocks for very high pack voltages

[jmmorgan]

I need help trouble shooting the Zephyr BMS lite charge controller. The kit was well packaged, but the instruction were very basic, but had good pictures. The assembly looked straight forward and was not hard. When I tested the unit the LED came on as described, but when the LED was green I still was measuring some voltage across the Pack pins. According to the instruction, I should not see any voltage. I have looked at the circuit too many times and I cannot see anything wrong. I elected to plug in the pack and see how the controller worked. Unfortunately, that was a mistake as the R1 resistor fried (the fuse was fine) and the controller no longer works, even when I replaced the R1 resistor. Any help would be greatly appreciated.
Thanks
John

[fechter]

I'll have to check those instructions. It would be normal to see some voltage across the pack connections even when the LED is green.

R1 is 1M and I can't see any way that should ever get enough voltage to burn up unless you're running over 300v. I suspect the wrong value was used for R1.

I'll PM you for more details.

[SlyCayer]

Hey guys, any news on the new version?

[dfar]

I am also anxiously awaiting news of the Zephyr BMS.

[chroot]

... bump

[SlyCayer]

Hmmm, this looks completely dead or it's only me? .

[fechter]

Sorry guys....

I'll try to at least get you an update on what's going on.
Basically both of our wives have prohibited us from working on the project anymore after wasting so much time and money with little hope of recovering it.

I'm literally working 3 jobs right now to make ends meet with 3 kids in college. I don't have lots of spare time these days.

I'm trying to work out some kind of alternate plan since the design seems to be finally worked out. It would be a shame to put so much into the thing and let it die at this stage.

[dnmun]

don't be sorry, not for the tourists anyway.

you and gary did a great thing, i still think this is the best thing to come out of the sphere. a lotta people know how much you guys have done.

so elegant and capable design, i was amazed you guys could devote as much time to it as you did.

really, you cannot imagine how the work you have done over the years has inspired me to learn some electronics finally in my old age. from the original hacking of the C'lyte controller, the puma/infineon solution and all the time you guys devoted to making the premier BMS solution. not once but twice, and all of it open source.

you guys should be proud. not sorry.

maybe some of the tourists will finally wake up and start studying electronics enuff to help out instead of whelping.

[megacycle]

I'm fairly a newbie, with no cred here, but If the membership could send you a dollar or more 10K+.
Need to have a whip round to get you home.

[chroot]

... RIP Zephyr...

Otherwise I understood your situation.

[fechter]

... RIP Zephyr...

Otherwise I understood your situation.

Not quite dead yet, just sleeping.
If I could find a source of funding, I could possibly get some bare boards made.
They won't be cheap though.

[Alan B]

PC Boards $5 per square inch for three boards:

http://dorkbotpdx.org/wiki/pcb_order