Home grown BMS ideas !

[weber]

My voltage divider ratio will always be less than one as it is a divider!
...
Perhaps "division factor" would be more exact than "division ratio".

The thing you subtract one from is the answer to the question, "If it's a divider, what's it dividing by?". Division factor is a fine name for it. Clearly your "division ratio" is just the reciprocal of this division factor. [Edit: a number you're dividing by is also called a divisor]

Take R2 as the top leg and R1 as the bottom leg, then the division ratio, Dr = R1/(R1+R2). A bit of algebra and re-arranging gives me R1/R2 = Dr/(1-Dr). So if my Dr is .63 say, then the resistor ratio is .63/(1-.63)= 1.702. from your table, the nearest ratio is 1.697 or 56/33. Putting R2=33 and R1=56 gives an actual ratio of .6292 which is pretty close.

So what have I missed here?

No need for anything as complicated as that. When designing, you typically start with two voltages. So instead of calculating Vout/Vin (small/big), you calculate Vin/Vout (big/small). Then you can "merely subtract one" to get the resistor ratio to look up in the table.

[Nevilleh]

I used a circuit very much like this a while back and one thing I noticed was that the voltage drop across the battery straps / terminals (particularly by the fourth cell) needs to be corrected for when the cells are under high load (3 to 5C).

Yes, one tends to regard copper straps as zero ohms, but they do drop a little voltage and the 4th cell gets the drop across 3 straps. I was thinking of using Al for the battery straps too.
I also wonder if it is of much concern to measure the cell voltage very accurately when they are under a heavy load? The point of the BMS is to ensure that no cell gets too high a voltage across it when charging and no cell reaches too low a voltage when discharging. So if your max discharge current is say 5C and you know the cell's internal resistance you should be able to come up with a Vmin figure that allows a bit of a margin. I note from the discharge curves that most of the cell capacity has been used when the terminal voltage drops to 3v (at no load) so if you are sucking 500A and the esr is 1 milliohm then 2.5v is the "alarm" point. If you drop the load, the cell voltage should come back to 3v and so the cell has not been discharged too much. I think the no load voltage gives the state of charge and provided it is above 2.5v then the cell voltage can fall below this due to the load current drop and that should not harm the battery as long as the no load vo;ltage can still come back above 2.5v.
This is still speculation on my part, of course!

[coulomb]

I was thinking of using Al for the battery straps too.

Why is that, Neville? Are the copper straps rather thin? Remember that only silver conducts better than copper (for the same cross sectional area), so you'd want to use thick aluminium straps.

I also wonder if it is of much concern to measure the cell voltage very accurately when they are under a heavy load? ... I think the no load voltage gives the state of charge and provided it is above 2.5v then the cell voltage can fall below this due to the load current drop and that should not harm the battery as long as the no load voltage can still come back above 2.5v.

Interesting point. I think I read somewhere that a real SOC measurement (taking into account the Peukert effect, less than for lead acid but still present) is better than using the terminal voltage. But how do you "debug" your algorithm without losing a lot of cells?

The other thing is that open circuit voltage may not spring back to a decent "state of charge" value immediately. Lead acid takes at least an hour to settle down, but the chemistries are different enough that this may not apply here.

The limits, either too high or too low a voltage, are very likely to happen at maximum charge or maximum discharge respectively, so I think it's reasonable to assume this and take typical voltage drops into consideration. Internal resistance varies with many factors, so I wouldn't take it too far; best to be conservative with cell protection.

[Nevilleh]

I was thinking of using Al for the battery straps too.

Why is that, Neville?

I had thought to use the Al heatsinks for the clamp as connectors. ie the +ve of one cell is connected to the heatsink via the pass transistor collector and it would be simple to extend the heatsink to reach the -ve terminal of the next cell.


Yes, actual SOC is difficult! It's easy enough to see how many AHrs have been used as we have a current measurement, but turning that into available capacity depends on estimating the load current.
I worked out the Peukert exponent for the LFP40 to be 1.0328 and I suppose one could take the average current used at any particular point and say that is likely to be the same for the rest of the cell capacity, put that value into the Peukert equation and hence derive a reasonably accurate figure for available capacity.

Nothing more embarrassing than running out of energy in an EV!

I have found with PbA in my electric scooter that even when the controller shuts down due to low batt volts, if you switch it off and wait a few minutes, it will come to life an struggle another 3 or 4 kms. Running the same machine on LFP40s is a different story. If you let them go down to the shutoff point, there is NO recovery, you have gone as far as you can go. I have 16 cells in series and I find that 44V is the minimum - if you aren't home by then you might manage another 3 or 4 kms if you drop down to 20 kph!

[Nevilleh]

My voltage divider ratio will always be less than one as it is a divider!
...
Perhaps "division factor" would be more exact than "division ratio".

The thing you subtract one from is the answer to the question, "If it's a divider, what's it dividing by?". Division factor is a fine name for it. Clearly your "division ratio" is just the reciprocal of this division factor. [Edit: a number you're dividing by is also called a divisor]

Take R2 as the top leg and R1 as the bottom leg, then the division ratio, Dr = R1/(R1+R2). A bit of algebra and re-arranging gives me R1/R2 = Dr/(1-Dr). So if my Dr is .63 say, then the resistor ratio is .63/(1-.63)= 1.702. from your table, the nearest ratio is 1.697 or 56/33. Putting R2=33 and R1=56 gives an actual ratio of .6292 which is pretty close.

So what have I missed here?

No need for anything as complicated as that. When designing, you typically start with two voltages. So instead of calculating Vout/Vin (small/big), you calculate Vin/Vout (big/small). Then you can "merely subtract one" to get the resistor ratio to look up in the table.

Yes, I see what you mean. But I have always thought of a voltage divider as something that reduces the voltage by multiplying it by a number <1 and my mental image is that of the bottom bit over the whole, if you like, as a factor. You obviously think of it as the whole divided by the bottom bit which is perfectly logical and gives a divisor rather than a factor.
The result is the same. Taking the .63 figure I used earlier gives your divisor as 1.587, subtract 1 to get .587, find 5.87 in your table (5.89 is nearest) to get resistor values of 330 and 56, now remember that you had to multiply by 10 and so one of the resistors is 10 times too big, so divide it by 10 to get 33.
I like my own way of thinking of it better! Understandable as I've been doing it for lots of years.

[woody]

From the Thundersky list:

Good technical article on Lithium battery management:

http://www.edn.com/contents/images/6648791.pdf

Managing high-voltage lithium-ion batteries in HEVs

SKYROCKETING ENERGY PRICES AND THE GROWING CONCERN OVER CARBON EMISSIONS HAVE FOCUSED ATTENTION ON ELECTRIC AND HYBRID-ELECTRIC
VEHICLES. New lithium -battery designs will be key technologies for efficient EVs and HEVs.

[coulomb]

In response to hemonster's query about my shift register idea, here are some thoughts:

   * Three opto-isolated daisy chains instead of 1. This costs more in optos (cheap) and in wire and/or connectors, but hopefully not much more in effort (maybe crimp multiple wires at once from ribbon cable).
   * Extra two opto-isolated connectors are clock and data for a CMOS shift register. I'm thinking two bits of data, so a 2-bit shift register per BMS board, giving 4 states: 0=normal, 1=read voltage, 2=read temperature, 3=lamp test. Might find a better use for code 3; I thought it might be handy to be able to get a particular cell's BMS to say "here I am" and test its LED(s) at the same time.
* The data is daisy chained, so in effect with 200 cells there is one 400-bit shift register.
* In normal mode, the cell puts "badness" on the bus, as usual.
* In read voltage mode, the addressed BMS reports its voltage in PWM, with a range of say 2.5 to 4.0 volts. This would obviously be fairly low precision, and relies on all the other cells being in zero badness mode. Or mode 3 could be "hands off the bus".
   * In read temp mode, the addressed cell would output a PWM width proportional to temperature, say 0-100C for min .. max PWM width.
   * Only one BMS would be addressed in non-zero mode at any time. Suppose you have 200 BMSs, and want to read the voltage of cell 102 (with 0 as the first one). You would send 204 (2x102) clock pulses, and data of 010000....0000, so the "01" gets shifted to the 103rd cell. It measures its voltage, puts it on the analogue output opto line, and the master BMS displays the voltage. To go back to normal, you output 400-204 clock pulses with all zeroes data. Or just 400 clock pulses to be sure.
   * This is relying on some 555-like device to translate voltage into pulse width. It would be rather non-linear, and the capacitors and resistors may drift with temperature and time. The non-linearity can be corrected for somewhat in software. I'd expect only about 10-20 "buckets" of voltage could be reliably distinguished; that could give the voltage to within about 0.1 volts, and temperature to within 5-10 degrees centigrade. It's not good enough to detect slight imbalances starting to happen, but should be able to tell you something like "cell 102 needs checking" and "this group of cells seems to get hotter than that group".
   * There could be fun with electrical spikes and clock skew, but if it's done slowly enough with proper shielding and capacitors everywhere and so on it should be able to be make to work reliably. At 9600 bps, 400 clock pulses are only 42 ms. Even at 1000 bps and 244 cells, it is still less than half a second. So speed should be fine to do "random access". Reading the voltage of all cells in sequence would be easy, as would be the reading all temperatures.

Comments welcome, as always.

Edit: 2 bits *per BMS*.

[coulomb]

Acmotor,

we like your BMS board, but we are wondering how you connect from one cell to the other. You appear to have resistors where some of the high current links/straps would have to go.

Do you intend to space the links off the PCB somehow to clear the resistors?

Go under the PCB somehow?

Do they kink up to clear the resistors?

Other?

I'm sure it's a clever scheme, whatever it is.

[coulomb]

A new consideration for per-cell BMS design, with the coming new NCOP version 2, is that cells will likely have to be restrained in their boxes. They will have to withstand 10 g in the vertical direction, so something probably has to bear down on the top of the cells.

For small cells like Thunder Sky 40 Ah cells, this is a bit of a challenge. I thought of metal bars bearing down on some 5 W resistors, but Weber thinks it's unlikely that we'll find resistors with mechanical stress ratings. I can't really argue with that.

Our engineer has okayed using 38 mm x 38 mm x 5 mm fibreglass L-shaped angle in place of 30 mm x 30mm x 3 mm steel angle, with a maximum of 20 mm holes in the fibreglass. (We thought the 20 mm hole would be big enough to clear the aluminium terminal nuts, but 25mm is actually needed).

So we're playing with the idea of having the fibreglass angle (with a 30mm dimple and channel routed in it for the vent) running between the terminals of each string, with suitable cutouts for the terminals. But then, the BMS PCB would have to go over the fibreglass, so it would be a PITA to get one cell out. The PCBs could plug in, but that's a source of cost and unreliability.

Clever solutions to this problem are most welcome.

[Johny]

Put a lid on the box? It could be an alloy lid with an inner Lexan (Polycarbonate Sheet) lining.

[coulomb]

Putting a lid on the box such that the cells can't leave the box, is acceptable, assuming you don't care about the cells in the event of a 10/20g event. At least, it should pass NCOP.

BTW, we found it difficult to find fibreglass angle etc. Here is a supplier that seems to have a presence in Melbourne:

Exel composites

[Johny]

I can just about go past their door on my way to work if I hang back a bit. I notice they also have a flat profile. I wonder if you could build and entire battery box with their stuff?

[Mr. Mik]

Hi,

I read through the thread and most of it is (so far) way too complicated for me.

But I have built and am using a Manual BMS (M-BMS) in the 102s NiMH battery of my Vectrix, it's homegrown alright, and might help to cross-pollinate some ideas for the more automated BM Systems you are discussing here.

The M-BMS works because I have capacity tested all cells individually and arranged the worst ones in order of capacity at the positive end of the string. Therefore I can quite reliably predict where I need to focus my manual monitoring efforts.

The M-BMS allows individual voltage monitoring of the 13 worst cells and of a few good cells (for comparison). It also taps into the 12 stock battery temp sensors of the Vectrix and allows monitoring of temperature at 6 additional locations in and around the battery.

It allows individual recharging of all 13 weak cells, either with auxiliary battery packs during riding or parking, or with an external NiMH charger that can charge from a single cell up to 18s.



Schematics for the entire M-BMS and for the Rotary switch Array can be found on Endless Sphere:
Vectux M-BMS 2009-05-10
Vectux Rotary Switch Array 2009-05-10



(After removal of the battery housing cover) the M-BMS also allows for charging of all individual 8s and 9s modules of cells with an external NiMH charger. That could be useful after seasonal breaks, or repairs, to balance the string prior to charging with the on board stock charger.

Each added tab consists of a 20A fuse directly at the cell, then 30A rated cable to the top of the battery pack. Connectors on top of the pack allow for the 8s and 9s module charging or discharging, and divide the tabs into monitoring cables (for each tab, with 15kOhm resistors in line) and recharging/discharging tabs (for the 14 cells at the positive end of the string and tab 102-028, without resistors but with an additional 10A fuse within easy reach).

That allows safe monitoring of the full pack voltage of up to 153V (due to the resistors limiting current) and limits the maximum voltage at the low resistance connections (which are easily accessible) to a relatively safe 40V.


(Edit: Clickable links to .pdf files on Endless Sphere inserted)

[acmotor]

coulomb, re BMS PCB.
I am using 2 x 6mmsq wiring (good for 120A cont. (84kW) ) with 2 x crimp terminals over the TOP of the PCB.
These are low cost and light weight and flexible arrangement.
(Red Suzi uses 1 x 6mmsq for 60A (peak 80A))

You could still use straps under the PCB with washers between PCB and strap. Strap would normally be heatshrinked. required insulation voltage is only 3.3V nom.

[coulomb]

I am using 2 x 6mmsq wiring

Ah. They must be a fair but smaller than these lugs for 16mm² wire:



As you can see, two such lugs could short in the middle, and are way too long to connect adjacent cells. Edit: I guess they could connect at a very oblique angle.

So under your PCB where the straps would go is clear of components? In fact, is it completely clear of components under the board?

Will you be using washers anyway to clear the vents? You don't seem to have holes in the middle to allow the vented gas to escape.

Will you be removing the Thunder Sky logo thing that bulges a little above the terminal height?

Sorry for so many questions.

[Johny]

acmotor - did you really mean 6mmsq, not 16?

[acmotor]

There are absolutely no components on the bottom of the board and the power resistors (through hole) are off to one side.
I will leave the TS badge in place so will use 6mm (hole) x .9mm s/s washers as spacers under PCB.



There are a few vias not actually required that I would remove if I made PCBs again.

The big copper areas are for max heat sinking away from terminals.


Typically 6mmsq copper wire is good for 60A (carrying capacity determined by temperature rise)
16mmsq copper wire is good for 105A i.e. 2 x 6mmsq is better.
This is limited by cooling surface area.
Voltage drop is just related to sqmm.

typical data:
mmsq   g/m    nom A    fault A(1sec)   mV/A.m
6             85          60             792            7.2
16          183       105          2112            2.7

Based on Rodeo 22kW ACIM and 220 x TS 40Ah.... (average current 40A)

Voltage drop of wiring assuming say 15m (over estimate) 2 x 6mmsq wire is (15 x 7.2) /2 = 54mV/A so at 120 A = 6.48V in 650V =1% I can live with that given the voltage drop due to battery ESR is 120 x .004 x 220 = 106V !
Also the peak wire current capacity is > TS s/c capacity and several times the rupture amps of the 125A fuses.
Now all this produces some heat (80W at average current and 10 x that at peak current, but then Matt assures me the TS perform better when toasty anyway !)

So that is my thinking since the Rodeo is just Red Suzi x 2 and that works fine.
2 x 16mmsq would be more like it for the 240A you have in mind ?

[weber]

Sorry acmotor, but I was with Johny in wondering if you meant 16 mm^2 or perhaps 6 B&S(AWG) which is 13.3 mm^2. I agree that voltage drop isn't an issue with even the longest runs in an EV when you're dealing with 600 V. But I can't agree with your continuous current carrying capacities (CCC).

In AS3000 Appendix C, the conservative no-brainer rating for 6 mm^2 is 25 A.

To determine a less conservative CCC you need to know not only the conductor material and its cross section, but also the type and temperature rating of the insulation and how many other conductors it is grouped with, whether it is sitting on any surface, and how it is enclosed.

If we assume the most common insulation, which is PVC rated to 75°C, with two wires side-by-side, not touching any surface, in free air. Then my Pirelli cable guide (which claims to be based on AS3008.1.1) says 49 A for each 6 mm^2. As soon as you put them in conduit that drops to 38 A each.

If you go to a non-PVC insulation rated at 90°C or better, such as cross-linke polyethylene (XLPE) you can get 60 A each in free air and 47 A in conduit. With teflon or silicone rated at 110°C you can get 73 A each and 58 A in conduit.

Assuming your 6 mm^2 cable does have a 60 A CCC in this application, and you have two of them so that's only 120 A, why then are you using a 125 A fuse?

You will probably get away with two 6 mm^2 for intercell links because they are short enough to be heatsunk by their crimp lugs and the cells themselves, provided the cells are not also very hot, as they might be in a fault situation. But you wouldn't want to be using only twin 6 mm^2 in longer runs with a 125 A fuse.

I wouldn't be using anything bigger than an 80 A fuse, preferably a 63 A fuse, with twin 6 mm^2. I think a 63 A fuse should tolerate 120 A for about 100 seconds and 240 A for about 25 seconds.

I also think your 1 second short circuit ratings must be assuming 90°C XLPE insulation or better. However I still agree they will probably withstand TS 40 Ah short circuit current, which I'm taking to be 1200 A when the cells are hot.

The rule of thumb I use is 100 ampere root-seconds per square millimetre. So 12 mm^2 of copper with 75°C PVC insulation should stand 1200 A for 1 second or 240 A for 25 seconds. That allows for the wire to already be at 75°C.

[acmotor]

Firstly, if my 125A fuse (Rodeo) does not blow in 1 second with the TS40 peak current flowing through it then I'll buy a lottery ticket ! We are talking HRC semiconductor rated fuses here. Not BS88 fencing wire !
Fusing is intended to protect against s/c conditions. Overload is controlled by VFD. Otherwise you must design wiring cont for max battery current !

The wire cont amp rating is so temperature and enclosure dependent as you note. Whatever number you chose, my cont current is 40A (22kW). This is below even the most pessimistic of the predictions.
My wire data says 85°C open air for the table in my last post.
You did get the point that 2 x 6mm are better than 1 x 12mm when it comes to CCC ?

Keeping in mind that your power expectations may be more than mine !

It is good to see that you have convinced yourself that my arrangement is neither over nor under-engineered.

[weber]

Firstly, if my 125A fuse (Rodeo) does not blow in 1 second with the TS40 peak current flowing through it then I'll buy a lottery ticket !

I expect it will. Who said it wouldn't? By "peak current" I assume you mean short circuit current, not the 10C pulse rating the manufacturer gives. Something like 25C to 30C when hot.

We are talking HRC semiconductor rated fuses here. Not BS88 fencing wire !

Good point. I wasn't thinking semiconductor fuse when I estimated those 100 second and 25 second overload ratings. But that doesn't affect the continuous rating.

Fusing is intended to protect against s/c conditions. Overload is controlled by VFD.

I'm not yet convinced of that. I don't think AS3000 would consider that sufficient overload protection. I don't think semiconductors are acceptable for that. What protects the cables from an overload due to a drive fault? Surely your VFD manual recommends fuse sizes and cable sizes. What does it recommend?

Otherwise you must design wiring cont for max battery current !

Not sure what you're saying here. You would only need to design wiring for battery s/c current if you had no fuses (or CBs etc), which would obviously be ridiculous.

The wire cont amp rating is so temperature and enclosure dependent as you note. Whatever number you chose, my cont current is 40A (22kW). This is below even the most pessimistic of the predictions.

Agreed. It's the 125 A fuse rating that I'm questioning now.

My wire data says 85°C open air for the table in my last post.

That's good. But still won't give 120 A CCC through 2 x 6 mm^2 when in conduit. And I wouldn't want other folks to assume that any old 6 mm^2 wire can do 60 A continuous.

You did get the point that 2 x 6mm are better than 1 x 12mm when it comes to CCC ?

Oh sure. Well aware of that. Was considering doing it myself.

Keeping in mind that your power expectations may be more than mine !

Yes, I have been keeping that in mind.

It is good to see that you have convinced yourself that my arrangement is neither over nor under-engineered.

Except for that fuse. Why do you think you need a 125 A fuse? Can you point us to its time-current curves?

If you need that fuse in order to tolerate your peak current requirements then I think you need at least 2 x 10 mm^2 wire, at least where it runs in conduit or other air-convection-restricted area.

However one thing I haven't taken into account is that with an EV you may never really get to a continuous rating for cables and fuses since the batteries typically last less than 90 minutes at max cruising speed. But why engineer it so close to the bone.

[acmotor]

125A (semiconductor rated)DC bus fuse is VFD rec. size.
Cable max size only is rec. (so as to fit in terminals)
VFD works normally (and current limits) or fails O/C or S/C. There is nowhere in between that can exist for any length of time i.e. a low resistance fault will progress to O/C or S/C in a short space of time. 125A fuse will blow. 2 x 6mmsq will carry fault current OK.

Remember the VFD is an electronic overload protection device in itself. It is superior in current limiting to any form of TCB, MCB or fuse.

I think the average currents in battery wiring in an EV (AC at least) are actually quite low. Check out the logged data on the red suzi post.

[commanda]

Hi there,
Not been on this forum for a while, so a bit late to the party. I have owned an EVT4000 electric scooter for 2 years. It was converted to Lithium 9 months ago, using 16 * TSLFP40 and a home grown BMS. I did a full write up here;
evt 400e lithium conversion thread

There is a pdf of the circuit there. It was rebuilt after the last post, and has been working perfectly the last 7 months. Of particular interest may be the use of constant current sources, especially for the final balance stage.
If I was building this at a larger scale, I would use current limited power supplies for the bulk charging stage, and change the first CC source to a simple switch.

Amanda

Mod EDIT: made link "clickable"

[Nevilleh]

Hi there,
Not been on this forum for a while, so a bit late to the party. I have owned an EVT4000 electric scooter for 2 years. It was converted to Lithium 9 months ago, using 16 * TSLFP40 and a home grown BMS. I did a full write up here;
evt 400e lithium conversion thread

There is a pdf of the circuit there. It was rebuilt after the last post, and has been working perfectly the last 7 months. Of particular interest may be the use of constant current sources, especially for the final balance stage.
If I was building this at a larger scale, I would use current limited power supplies for the bulk charging stage, and change the first CC source to a simple switch.

Amanda

Mod EDIT: made link "clickable"

I looked at your link and found it quite interesting. I have been working on a BMS for a little while and what I have is surprisingly similar to what you wrote up, although uses a lot fewer parts. Perhaps not so surprising when you consider we are after the same end result.
Mine consists of a shunt for each cell, similar to yours, but I used an Atmel Tiny25 micro to measure the voltage of each of 4 cells (and the temperature) and send the data via an optocoupled link to a "master" micro that outputs signals for "cell low", "one shunt on", "all shunts on". It also has a "High temp" output. Four such micro pcbs cover 16 cells.
The "cell low" signal can cut off the load or simply alert the driver to back off the throttle and head for a charging station.
The "one shunt on" signal shuts the charger down to 1.5A so the shunt can carry the charging current without getting too hot and the "all shunts on" signal indicates that charging is finished.
I used an EVT4000e scooter to check it all out with! Small world.
My charger comprises a couple of nominally 27v SMPS in series adjusted up to 58.4v (3.65v/cell) total with 6A current limiting and I hacked them to shut down to 1.5A when provided with the appropriate signal - "one shunt on". They also have a shutdown input which is activated when charging is finished.
I plan on extending the system to 45 cells for my EV project and that will also require a much bigger charger as they are 120 AHr cells and I want to charge them at 20A or so, the highest power I can draw from 230v, single phase mains.

[woody]

I had a simple idea last night for attaching a BMS board which manages a group of cells (e.g. 16-20) without rigidly attaching to the cells, and without fly wires.

1) Use bent copper strap in a upward bridge shape to join adjacent cells as per a4x4kiwi
2) have a hole in the top of the bridge
3) mount the PCB to these holes
4) 2 extra straps for POS + NEG end or the pack.

This gives the BMS board:
a) access to each end of each of the 16-20 cells
b) a semi-rigid mounting platform
c) only n+1 connections
d) more board real estate per cell compared to usual "per cell" design.
e) approx same PCB area (cost) as per cell designs.
f) cheap + quick connections

Our art studios have developed this concept drawing:


Other ideas built on this platform:
i) Mount the reversing contactor to the BMS board
ii) Mount the heatsinks in a row and have a fan or 4 with ducting to move air directly across the heatsinks.

thoughts?

Woody

[weber]

Great idea Woody! All the parts could be underneath for mechanical protection and to minimise height. And if that includes bypass power resistors and transistors then it should also include holes directly above those power dissipating parts. Perhaps temperature operated fans can be mounted underneath too (several very small ones).

As with the connecting links themselves needing wiggles for stress relief, I worry about cell swelling and other movement stressing the PCB connections.

One good thing about it, that was brought to my attention recently, is that it avoids clamping circuit boards under the terminal bolts involved in the high-current connection. Bolting PCBs to terminals seems to be asking for trouble (sorry acmotor) because of the phenomenon of "creep". The epoxy that the PCB is made with will slowly relax over time and then the connection will become higher resistance and get hotter and the creep rate will go up in a vicious cycle until something catches fire. Yes there are copper vias in there, sort of like tubular copper pillars, but they are only thin walled and may buckle as the epoxy relaxes.

Perhaps such boards could be saved by drilling, holesawing or routing out the PCB at the terminals and inserting tinned-copper crimp-type rings and soldering the crimp part to the PCB.

BTW, acmotor, do you want us to do any more cell tests with your BMS in place, or should I mail it back to you now? Thanks for the loan of it.