Designs for Reliable (low V) Cell Monitoring electronics

[CamLight]

Any comments on using DB9 connectors for this application? They are available in pc board mount, good for many cycles, and commercially made pre-molded cables are readily available to cut up for the battery side, so all that need be done is sort out the wires and crimp ring lugs on (for Headways at least). No special crimper is needed.

IMHO, DB9 connectors are great. They're fairly rugged when connected, widely available, and inexpensive. But, the connections are wide open to the environment. I guess if the monitor is inside whatever weatherproofing the pack has then sealed connectors aren't required though.

Do we want a PCB-mounted or cable-mounted connector?
PC-mounting it makes it very hard to replace and prevents the waterproofing of the PCB. Coming off the PCB with a cable to connect to the pack's cable allows for more inexpensive options for wrapping/encasing/protecting the PCB. And replacing the connector is easy if the cable wasn't set up to a just-barely-long-enough length at the start.

But, a PCB-mounted DB9 does make assembling (and stocking, if anyone is selling it) the board easier. Guess that has to be weighed against the longer-term considerations.

 

[Alan B]

For reference and possible discussion, here's a micro-per-cell design that worked beautifully for me. I easily achieved +/-10mV cutoff accuracy without calibration from 0C to 40C. It's easily scaled down as most of the components can be removed to tailor the feature set.

If an ADC is available in a tiny micro (6-8pin) you'd like to use, then you can remove the op-amp and supporting circuitry. If the micro has an accurate enough internal voltage reference to meet the requirements, you can remove the LM4050 too.

Just tossing this into the mix for comparison and possible ideas to borrow. Scroll down to the third article in attached PDF.

Nice design. About 22 parts per cell. The cell select line was interesting. They did a nice job of turning things off to save power. Does that PIC have enough clock accuracy for serial comms, or perhaps they had to calibrate that in some way. As you indicated it would help to have an ADC that didn't need so many external parts. A resistive divider could also be turned off with a binary output pin so 2 pins with a good internal ADC would do the trick. Also if the micro's internal temperature sensor could be used that would save another part.

I made some rough sketches with a single AVR per cell. Not quite as many parts, still a lot compared to the 8 cell design. However these micro per cell designs are full BMS solutions, not just voltage detection.

The current eight cell monitor design is about 43 parts for 8 cells, or less than 5.5 parts per cell.

 

[Alan B]

Occasionally I wonder if a switched capacitor voltage sampler would be good for this monitoring application. It would be low power and more accurate but would add the switching matrix. Have to think about that one.

 

[dak664]

I made several 12 cell monitors using an ATTiny45 for each cell, a breadboarded version is here http://endless-sphere.com/forums/viewtopic.php?f=14&t=13717
They broadcast through RC5-ish infrared, to an ATTiny85 receiver powered by USB. They sleep at 1 microamp until hit with a reset pulse that is capacitively coupled from the previous MCU. The first MCU wakes every 8 seconds to initiate the cascade, so pulling it from the socket effectively shuts the rest down.

If the first MCU used an external trigger then there would be practically no load on any of the cells unless the triggering circuit was active. You could have an on/off switch, or powered only when USB is connected, or power it from a smaller battery that would only last a year so if accidentally left on in storage. After 6 months it could start with the beeping and blinking.

 

[CamLight]

Nice design. About 22 parts per cell. The cell select line was interesting. They did a nice job of turning things off to save power. Does that PIC have enough clock accuracy for serial comms, or perhaps they had to calibrate that in some way. As you indicated it would help to have an ADC that didn't need so many external parts. A resistive divider could also be turned off with a binary output pin so 2 pins with a good internal ADC would do the trick. Also if the micro's internal temperature sensor could be used that would save another part.

I made some rough sketches with a single AVR per cell. Not quite as many parts, still a lot compared to the 8 cell design. However these micro per cell designs are full BMS solutions, not just voltage detection.

The current eight cell monitor design is about 43 parts for 8 cells, or less than 5.5 parts per cell.

I was up at 9600 baud using the internal clock on the same PIC for my first prototype without problems. I didn't need to go any faster so never tested it at higher baud rates. A crystal (and 2 caps) can always be used instead for much higher baud rates (because of increased accuracy) but that ups the parts count. But if you needed higher baud rates, the parts count increase really wouldn't matter.

What kind of baud rates can you achieve using the other processor families (on their internal clock)? Not saying to use PICs for the design, just technically curious.

A monitor-only version of the micro-per-cell circuit would probably be about 10 parts? Would have to weigh the cost vs. advantages/disadvantages of the features available in each design. IMHO, once we're down to these relatively low part counts, comparing the counts probably doesn't matter as much as the features delivered by the design. And, perhaps, total part cost (but I think that will be pretty close).

But, unless there's a specific feature/spec in another design that is a requirement, your shared micro design is going to be hard to beat.

 

[texaspyro]

However these micro per cell designs are full BMS solutions, not just voltage detection.

Yeah, but once you can monitor the voltage on each cell and have some smarts floating around, you are a pretty small step from something really useful... a real BMS. Just having a LVC (and HVC) monitor is not all that useful. I think the most useful thing one can have to keep an eye on the pack is a bar graph display of each cell. Impending problems show up quite well

You just need to control a honkin' low side FET switch to disconnect the pack if you see an LVC (mostly for discharging) or HVC (mostly for charging). Toss in a current shunt and you get full pack protection and the ability to do all sorts of Cycle Analyst stuff.

While you are at it, add the switched capacitor balancer to help keep the pack in whack. It's not that good for initially balancing a new pack (but should be able to do it over a weekend or so), but should help keep it from ever getting out of balance.

 

[Alan B]

I was up at 9600 baud using the internal clock on the same PIC for my first prototype without problems. I didn't need to go any faster so never tested it at higher baud rates. A crystal (and 2 caps) can always be used instead for much higher baud rates (because of increased accuracy) but that ups the parts count. But if you needed higher baud rates, the parts count increase really wouldn't matter.

What kind of baud rates can you achieve using the other processor families (on their internal clock)? Not saying to use PICs for the design, just technically curious.

A monitor-only version of the micro-per-cell circuit would probably be about 10 parts? Would have to weigh the cost vs. advantages/disadvantages of the features available in each design. IMHO, once we're down to these relatively low part counts, comparing the counts probably doesn't matter as much as the features delivered by the design. And, perhaps, total part cost (but I think that will be pretty close).

But, unless there's a specific feature/spec in another design that is a requirement, your shared micro design is going to be hard to beat.

The precision of the oscillator doesn't limit the baud rate, it limits the word length. The error accumulates over the bits. It can be calibrated, then the stability is the issue.

My micro per cell sketches are about 12 parts per cell. Plus the master. Leaving out the balancing doesn't help a lot, it is only 2-3 parts. These per-cell micro designs don't actually meet the requirements for surviving wiring errors. But they do provide some nice extra features.

 

[CamLight]

The precision of the oscillator doesn't limit the baud rate, it limits the word length. The error accumulates over the bits. It can be calibrated, then the stability is the issue.

You can also say "The precision of the oscillator doesn't limit the word length, it limits the baud rate." A very long word length is OK with the right baud rate (i.e., low enough). I haven't compared the stability of clocks of various micros though so I don't know whose is "best". Especially over the possibly wide temperature range the monitor might experience. But your design doesn't require serial comms so that makes it easier...no need to compare!

This question is mechanical-component related, and your thread is about reliable electronics, but they're intertwined pretty closely in this application so I hope you'll forgive me for asking....
Any limitations on the monitor's operating temperature in your specs or any requirements for wiring insulation temp. rating? IMHO, softened (if not melted) insulation has lead to lots of failures which might be avoided by not allowing operation above a certain temp or requiring higher temp. insulation?

 

[Jeremy Harris]

Chaps, I think you may be going down a rabbit hole (worrying about another non-problem.....) with the debate on frequency stability and baud rate/word length.

Given that the tiny amount of data that needs to be transmitted from channel to channel (or master) only needs to be sent/received very slowly (polling once a second is probably more than fast enough and you'll be sending what, two or three bytes per channel?) and also given that we used very reliable serial comms long ago with horribly unstable and inaccurate clock generators, I really don't see that there's a problem. I remember making a simple serial link 30 odd years ago using 555s as clock generators. It worked fine for transmitting data from a weather station into the house for years. Similarly, teleprinters worked very well (albeit at 300 baud!) with no fancy timing on the serial links.....................

Jeremy

 

[Alan B]

You can also say "The precision of the oscillator doesn't limit the word length, it limits the baud rate." A very long word length is OK with the right baud rate (i.e., low enough). I haven't compared the stability of clocks of various micros though so I don't know whose is "best". Especially over the possibly wide temperature range the monitor might experience. But your design doesn't require serial comms so that makes it easier...no need to compare!

This question is mechanical-component related, and your thread is about reliable electronics, but they're intertwined pretty closely in this application so I hope you'll forgive me for asking....
Any limitations on the monitor's operating temperature in your specs or any requirements for wiring insulation temp. rating? IMHO, softened (if not melted) insulation has lead to lots of failures which might be avoided by not allowing operation above a certain temp or requiring higher temp. insulation?

Here's how I think of it. The clock generator frequency limits baud rate. A high enough frequency is needed to generate the baud rate clock. Some generators have particular divider arrangements so they need specific clock frequencies to work with while others have programmable dividers that can be set for period rather than division ratio and those accomodate generating most any baud rate if the clock has sufficient resolution. The RC clock in the ATTiny 25 is 8 mhz.

The clock generator accuracy and stability limits word (really character) length. The asynchronous serial stream has a start bit, some data bits, possibly parity and a stop bit. The error accumulates over this length from the start bit. If the clock error is 10% (these internal clocks are RC and the uncalibrated spec is plus or minus 10%) on the receiver and 10% the other way on the sender then the total error could be 20%. At the end of the start bit the error would be 20%, at the end of the first data bit it would be 40% and the second data bit would be 60% which is enough to miss the bit and those thereafter. This is worst case.

If the built in RC clock oscillators are calibrated during programming they can be corrected to 1% but this involves extra steps in programming and some way of measuring the clock and computing the calibration factor. This accumulates to 10% over the transmitted character which is not a problem. Center sampling requires a precision of about 50% minus 1/16 or so depending on how the bit center timing is done.

If the chips are on the same board one could run a clock to all of them but it consumes pins on the chips and on these small ones there are few to work with.

On the wiring, it would be a good idea to have some guidelines about wire insulation and gauge usage. I'm not sure if the cause of melted insulation is the wrong insulation choice or the wrong wire gauge for the current. I see a lot of questionable wire gauge selection on ebikes, or just plain crazy current limit increases. Suggestions on what to spec or recommend here?

A temperature spec is a good idea. There's no real dissipation, but how hot do these black nylon rack bags get in the sun on a hot day? 130F?? The micro is rated to 125C!! RC Clock accuracy at 1% stability is rated to 85C.

 

[AndyH]

It'll be interesting to see how far 'out there' this goes.

Motor controllers are mini wide-band transmitters. The electrical noise alone can make digital devices useless even if the EMI/RFI doesn't kill the chips themselves. Then there's the usual other concerns like vibration, water, etc.

I have not lost one of Gary's analog BMS or LVC devices on the road, but have replaced a number of professionally-designed and manufactured uProcessor monitoring devices and uP-based BMS boards.

Andy

 

[fechter]

It would be fun to hook up a spectrum analyzer to the pack on an electric car or bike and see what it looks like under full load.

There are proven tactics to protect the MCU power supply against transients and EMI, but the signalling could possibly run into problems under some conditions. Again, proper design can avoid most of the issues, but may require more parts. One reason I'm a big fan of optocouplers.

 

[Alan B]

Here is the micro per cell BMS design that I was playing with some weeks back. It is not fully developed. It has about 11 parts per cell plus whatever is in the overall supervisor micro.

http://picasaweb.google.com/lh/photo/tChuueOJab_1R2SBxVpgOoPSvVPFxOvt9gQSg1eZ7bg?feat=directlink

I think the per cell design is interesting for more accuracy and information on each cell, but it is not as simple/robust as the design from post #1 of this thread. For example wiring it wrong to the battery will blow fuses or perhaps pop micro chips.

 

[CamLight]

On the wiring, it would be a good idea to have some guidelines about wire insulation and gauge usage. I'm not sure if the cause of melted insulation is the wrong insulation choice or the wrong wire gauge for the current. I see a lot of questionable wire gauge selection on ebikes, or just plain crazy current limit increases. Suggestions on what to spec or recommend here?

A temperature spec is a good idea. There's no real dissipation, but how hot do these black nylon rack bags get in the sun on a hot day? 130F?? The micro is rated to 125C!! RC Clock accuracy at 1% stability is rated to 85C.

The lowest temp rating I've seen for insulation is 85C so that might be a good start? It fits well as an upper limit for electronics and derating the insulation rating so inexpensive PVC wiring can't even soften would also help to set an upper limit for the pack (if the monitor is mounted against it). Perhaps 60C (140F) as the operating temperature limit for the monitor and 85C for storage?

We all know just about any limit would be instantly exceeded by any number of users , but at least the limit would be identified.

Hmm...if a certain insulation type for all wiring was a requirement (e.g., silicone or teflon) then 85C, or even higher, could be set as the limit.

 

[Alan B]

There are a few ways this three chip 8 cell monitor project might work:

1) pcboard only is made available (possibly with pre-programmed chip)

2) pcboard and parts kits are made available (perhaps with a pre-programmed chip)

3) built and tested units are made available

If it is done as boards or kits it probably must be through-hole.

If it is done as fully assembled then it probably should be surface mount (though it could still be through hole).

Kits may have calibration issues, have to sort that out.

We had one volunteer to lay out a circuit board, if I remember correctly. Do we have anyone interested in making and selling boards, kits or assembled units?

Are there any customers out there? If this gets built, how many should be made?? Is there any interest??? Should we start a separate interest thread????

 

[CamLight]

Are there any customers out there? If this gets built, how many should be made?? Is there any interest??? Should we start a separate interest thread????

I think this is a great time to do that!
Finalize the design, feature set, packaging/case style, component selection, probable board size, and the schematic and then estimate the parts cost. This will help everyone know how much the monitor might cost and, IMHO, help any potential customers decide if they're interested. I think cost will be everyone's first question in the new thread. Then you can work up a couple of prototypes and get them out for testing.

Personally, I feel that cost doesn't drop enough until you reach the 100pc. mark for the PCB and sets of components. I guess it all depends on the max. users would spend on the bare board and pre-programmed chip though.

I'd be willing to create and stock bare boards along with pre-programmed chips if there are a decent number of definite pre-sales. For me it probably wouldn't be worth stocking full kits as the ordering, kitting, repacking, documentation creation (instructions, BOM, etc.) and tech support time involved is considerable and really ups the kit cost. And I don't think I could even come close to making fully assembled, calibrated, and tested units affordable in quantities this low. Others may have more free time to help keep their labor costs lower though.

 

[Alan B]

OK. Where do interest threads go on this board?

I agree that 100 units is a breakpoint. Below that is quite expensive.

It also seems quite unlikely that anyone wants to sell built units. So surface mount is out.


A quick recap of this thread - The purpose was to design a more robust and reliable cell monitor. Requirements were evolved and a design developed. Other designs were analyzed and compared. The result is a set of requirements and a design that meets them. This significantly exceeds the robustness features of existing designs and deals with most all failure modes noted using fewer parts than previous designs and does so at low cost using readily available parts.

Eight Cell Three Chip Voltage Monitor:

Features, Draft 0.5:

8 cell capacity
Robust - incorrect wiring will not damage it (most common designs are damaged by even momentary wiring errors)
Protected against poor connections in pack (causes high voltages that damage most designs)
Pins can be connected in any order (many designs cannot handle this without damage)
Continues to work when up to 6 cells are at 0V (most other designs stop working when cells go to 0V)
Meets requirements in post #1 of this thread (no other designs have been presented or referenced in this entire discussion that meet all these requirements)
Fewer parts than most other designs (significantly fewer active components)
Transient protected, fuse on regulator
DB-9 male connector on PC board, cut and wire commonly available DB9 Female cable to battery
Easily disconnected from battery during storage for zero drain
Alarms on cell undervoltage
Alarms on cell overvoltage (and poor connections in pack when under load)
Continuously self testing - self test error generates alarm
E-brake output on alarm
Piezo beeper chirps on alarm to alert even when ebike is not in use
Red led blinks on alarm
Green led blinks when no alarm
0.1V accuracy
Draws less than 300uA (more when alarming)
Filtered and averaged to avoid false readings
Calibrate button compensates for component variations and sets to your batteries, LiPo or LiFePO4
Variable sample rate, twice per second when operating, every 10 seconds when idle
Through-hole kit, approximately 40 components
Covered with clear heatshrink for protection
Programming pads for firmware updates or software customization
Software source code available (free tools - GCC) for user customization

Cost - estimated about $25-30 for a kit. Solder in the parts, connect to batteries, push button to calibrate

Subject to change.

 

[Alan B]

Thinking about adding a calibration mode. Triggered by a pushbutton or jumper, reads the cell voltages and establishes the "normal" readings. Applies a window for over voltage and undervoltage to that nominal reading. This would make it easier to build as a kit, and would make it compensate for different cell types rather easily.

 

[CamLight]

I agree, IMHO thru-hole for everything.
And perhaps go to a (right-angle) DB-15 instead of DB-9 for the connector so the programming pins can be accessible from one end of the PCB assembly without removing the heatshrink?

For kit costs, also include the time needed to unpack and distribute the parts (including any necessary anti-static foam) into kits and also the time/cost needed for the kit bags, shipping boxes and bubble-wrap (or whatever). The cost and time might seem low per kit but adds up when, let's say, 100 kits are done. We're talking at least a weekend here. Unless the person selling the kits is supplying all of this for free, along with the time for kitting the parts, the kit cost will have to include this.

And, unless they're willing to donate the time and shipping, we might also have to include the cost for someone to purchase and program 100 micros and then ship them to the person supplying the kits?

 

[Alan B]

Good suggestions. The pricing is very rough. Need something to talk about.

I was also thinking about DB25. That would allow not only programming but also the ebrake lines to come that way.

 

[Alan B]

Just added a calibration button to the schematic. (The version I'm working on, not the one on the web.)

I'm thinking the best way to calibrate is to do it at "normal" resting voltages. Not hot off the charger, those vary too much.

Set the upper limit to something like 0.6 volts above nominal, and the lower to about 1.0 volts below nominal. These can easily be changed in software.

So for the Headway, with nominal at 3.3V the upper limit would be 3.9 and the lower 2.3 volts. Perhaps the lower should be a bit lower yet.

 

[texaspyro]

I've found the little voltage references at http://www.voltagestandard.com to be quite handy. For what they do, they are dirt cheap and very stable. Plus he will recalibrate them at any time for a pittance.

 

[AndyH]

Are there any customers out there? If this gets built, how many should be made?? Is there any interest??? Should we start a separate interest thread????

Grab a copy of "The 4-Hour Work Week" by Timothy Ferriss for a great step-by-step look at product development and test marketing. It's usually better to find an unmet need and willing customers first, THEN develop a product that meets their need.

 

[Alan B]

Hi AndyH. I have not read that book. Sounds interesting. I may look for a copy.

I believe that is what we did here. These requirements were built from experiences here and elsewhere with real failures and real applications, the product designed (and evolved) to meet the requirements, alternative designs evaluated, and at this point it is time to gauge customer interest before making the investment in the final development. The design will continue to evolve as the factors relating to production are folded back into the product. A significant amount of work was done before this thread was started, so it may not seem as "linear" a process. There was more overlap and more iterations occurred, but often these iterations produce a better product than a linear process.

There seems to be some disagreement here as to whether the existing designs are "good enough". At the same time some folks on here don't use protection because the existing designs "are not reliable enough". Many have experienced failures. I said at the outset that we could make a more "robust" design. I think we have done that. Perhaps it is "too good". Some claim that "micros are not reliable", but these are not folks who have designed reliable systems with modern parts, and the industry does not agree with them. Some claim that the parts reliability is not the problem, it is the interconnects or the environment. We have designed using better connectors and circuits that have very high resistance to damage from the environment. I think we answered all the issues, aside from executing an actual prototype. If there is no interest in product then perhaps it stops here.

On the voltage standards, I have a similar one. Pretty amazing devices and not terribly expensive. It is fun to put it on a quality DVM and read 5.000 volts.

 

[texaspyro]

On the voltage standards, I have a similar one. Pretty amazing devices and not terribly expensive. It is fun to put it on a quality DVM and read 5.000 volts.

I also have a Fluke 732B (and a few 732A's) and 720A Kelvin-Varley divider. The 732B reads 10.000000V. The 720A can divide that by down by up to seven decades. Only a super-conducting quantum Josephson Voltage Standard can do better. Not too shabby what you can score on Ebay with a little patience. I got both, and a ESI SR104 10,000 ohm standard resistor for less than what a single calibration of just one of my HP3458A voltmeters would have cost. But then, one probably needs to pay to have the 732B and SR104 calibrated (but I seem to be able to do that better than any of the local cal labs).

My SR104 is over 40 years old and pretty much all the long term drift has aged out of it. I have it in a very sophisticated environmental chamber (aka Coleman cooler) where its temperature is stabilized at its optimum calibration reference temperature to around 0.001 degrees C (with an RMS long term temperature stability of a few micro degrees). It's pretty amazing what one can do with some 50+ year old technology coupled to an AVR chip.