Build your own BMS

[s/v Jedi]

Originally Posted by Lake-Effect

I'm following this with interest. (I'm an ESP8266/ESP32 fan)

I have one question. This thread is mostly implementation details; is there some site or document that lists out the Li "rules" you're implementing? In other words, the list of ideal conditions/methods for charging and using Li batteries, and the conditions/tolerances that should be measured and alarmed for. And what's under control of the BMS.

Many of you already have Li banks so you've no doubt become very familiar with those rules, but it would help the rest of us keep up, and to learn more about using Li batteries. It would also be a scorecard for assessing different BMS systems, whether DIY or commercial.

You've possibly already done this in other CF threads; I'd be grateful for a link if so. Thx.



Yes, that seems like the direction to go.

I think most of that info is in my old thread from when I prototyped it: https://www.cruisersforum.com/forums...ms-230266.html

[goboatingnow]

The Rec bms manual gives a huge comprehensive spec as To all the set points very useful

[goboatingnow]

By the way reviewing TIs power monitor range of which the INA226 reveals one unfortunate drawback.

The adc can not be used to measure differential voltages over 80mv , only the single sided input can measure full range voltages with respect of chip ground.

It would be Better to read each cell in the string as a difference rather then ground referenced

Of course a precision voltage divider could be deployed to bring the cell voltage within the differential range

[s/v Jedi]

Originally Posted by goboatingnow

By the way reviewing TIs power monitor range of which the INA226 reveals one unfortunate drawback.

The adc can not be used to measure differential voltages over 80mv , only the single sided input can measure full range voltages with respect of chip ground.

It would be Better to read each cell in the string as a difference rather then ground referenced

Of course a precision voltage divider could be deployed to bring the cell voltage within the differential range

Read the original yaBMS thread on why that isn’t a good option and how, in the end, the INA family is the best option. The INA226 has a 16-bit ADC or 0.5mV accuracy. This may double when subtracting the reading from the previous cell, so 1mV. If you need higher precision then there are 20-bit versions, I think the 229.

[wholybee]

Originally Posted by goboatingnow

So a bms built by a time served industry professional with ieee credentials overa “ ONHUNGLO” bms brand

Sure right on
But you’d find sailors are only “hobbyists “ don’t go sailing unless you a fully certified mca Yachtmaster . I mean mere hobby sailors can’t be insured can they.

Some insurance companies have ridiculous rules, forbidding DIY installs, requiring ABYC certified techs install either Drop Ins, or U.S. made batteries and a BMS that is U.S. made, etc.

A DIY BMS, or even a DIY battery with a REC BMS, would not satisfy the stupidity of these insurance companies. I'm not going to let that force me into an inferior installation. There are other insurance companies out there that will insure you, and more and more, people are dropping insurance altogether.

[T1 Terry]

Originally Posted by goboatingnow

Victron does so at high Victron prices and you have to use their batteries , no thanks





Whilst I agree I feel for the tiny additional cost 3x 4S with every cell monitored is the safest possible system. No hidden parallel unmonitored cells.


I’m using LFp at the cheaper end of the price scale so more monitoring the better



I do not intend to balance between batteries as I see no point.

With 5A switched capacitor balancing I can trigger balance once the overall charge current has fallen near this figure irrespective of actual balance set point. In essence doing mid point and top balancing.

With this balancing , balancing can start earlier as no charge energy is being wasted. The higher cell energy is transferred to the lower cells. In essence once a “ imbalance “ threshold is triggers active balancing can be invoked irrespective of where on the SOC scale you are on.

This means balancing is better , with no energy loss and little delay in charge completion time. The passive system has little to recommend it.






As I said I’m using cells on the cheaper end of the scale

I’d prefer to monitor the cell voltage and temperature of every cell as well as the current of each “ battery “ ie the 4S string

Secondly the 3x battery ( to get my 12v 300 AH) mirrors the drop in approach but with proper network BMS each battery BMS is working with each other so a shutdown in one triggers all shutdowns. But with manual override to bring the pack back as a 200Ah or 100Ah capacity.

It also allows individual cell replacement while retaining battery functioning.

As for load sharing. I’ve had 30 years in industrial electronics including military and space based battery system. In this regard I can attest to a very long time of experience.

I’m not comfortable with load currents in excess of the individual cell. Cascade failure is a real issue in parallel systems.

In my case each “ battery “ consisting of 4S cells of 100Ah capacity will have battery based current protection.

If that battery current trips the BMS will shut down every battery until manual overrides can be triggered to decide what is appropriate to reconnect. This means system current is limited to individual battery max. I feel strongly this limits cascade failures.

I agree that LFP is reliable , I find some of these debates building all sorts of redundancy designs switching in SLA , bypassing this that or other rather “ amusing “ as if LFP failure which is , as you say, highly unlikely , is somehow going to happen every time the boat senses a Lee shore at night. Over protecting one unlikely failure mode while ignoring everything else is silly.

There seems to be a body of opinion that this “ new fangled “ “ Lithium thingy” is dodgy, prone to failure or burns the boat down whereas the truth , as you say it , is it’s actually more reliable.

However as an engineer I never take “ it’s worked for me for 11 years “ with any seriousness , experience has thought me proper engineering analysis is always better then specific on the ground reality ( cause the next guys “ ground” may be different )

My view is the future of Li on board has to be drop in batteries with networked BMS. this needs industry standardisation. But I’m going to build my BMS and packs on that basis.

I'll work from the bottom up,
However as an engineer I never take “ it’s worked for me for 11 years “ with any seriousness
I should have said I have a few hundred systems out in the field over the last 11 yrs. Basic engineering analysis is based off what worked in real life rather than theoretically should work but I'll add a 50% safety factor just in case (I can't find the :WINK: icon so read it as having been added here :lol

I’m not comfortable with load currents in excess of the individual cell. Cascade failure is a real issue in parallel systems.

In my case each “ battery “ consisting of 4S cells of 100Ah capacity will have battery based current protection.

If that battery current trips the BMS will shut down every battery until manual overrides can be triggered to decide what is appropriate to reconnect. This means system current is limited to individual battery max. I feel strongly this limits cascade failures.

As you have already posted, the LFP cells are reliable, the cheaper end of the scale just has less capacity to handle higher loads because the real capacity is a fudged figure, generally with the "you can only have 80% SOC " so in reality the battery is actually only 80% of what they advertise on the case. The next most common fudged technical bit is the load that capacity was tested at, and this is the true load limit in any single cell ..... advertised 100Ah capacity does not say you can pull 100 amps at any time and unlikely for an hr to get the full 100Ah of advertised capacity.
As an example, often the lower end of the market like to use the more common C20 rate for load testing and draw the battery down until the voltage drops like a stone .... and this is the point they measure the capacity. C20 for a 100Ah battery is only 5 amps load, if it is fudged "80%" figure, then the C20 rate is only 4 amps.
If you want to use the C1 rate as the base for max load, you need to test just how many Ah you get out before the cell drops below 3v while under load, then reset the cell capacity and C1 load from there. In the end, if you do the sums as to $$ per usable Ah, the Winston cells aren't all that expensive.

I commend you for attempting to make the drop in market a bit more reliable, it certainly isn't at the moment ....... but I still make a few $$ pocket money trying to at least get them to work together .... and that is the hard bit, full load is fairly easy, it's that small parasitic load that causes the hassles, a 1 amp load over 24hrs will pull the battery with the least total circuit resistance down to 75% capacity, yet the voltage will remain the same across all the parallel batteries .... you understand accumulative resistance so it doesn't take much to realise why one battery will end up supplying all the light loads for at least 50% of its capacity, yet a balancer that relies on voltage to sense where the capacity needs to be drawn from and where it needs to be sent to will not rebalance the the batteries, only cells in parallel and those groups in series can do that .....

T1 Terry

[flightlead404]

This is a great thread and I'm enjoying reading it, thanks Jedi for starting it.

I've got a fair amount of experience with Arduino over the last decade and I'd like to add one thing I've learned. The el cheapo Arduino you get on Amazon or ebay may be fine for experimenting, but when it comes to building something permanent, especially something as important as a BMS, I'd recommend going with the name brand real Arduino.

I've lots of experience with the cheap ones, and I've found they tend to fail. Sometimes physically, eg component soldering or traces, other times for no apparent reason.

[goboatingnow]

My own view is that 12 bit differential cell voltage measurements are better then a ground referenced single ended 16 bit ina226 approach

Hence my preferred approach will be a high common mode range unity gain instrumentation amp feeding the esp32 internal 12 bit SAR ADC. This has 6 channels ( io available) or I can use an enable feature to select a cell.

This will give me +- 1mV if I can keep the noise floor down

Overall current measurement will done by calibrated mosfet Vds as this is quite stable avoids shunts and the disconnects will be mosfet based but the overall bank will be also protected by a contactor controlled by all BMS

[goboatingnow]

Originally Posted by T1 Terry

I'll work from the bottom up,
However as an engineer I never take “ it’s worked for me for 11 years “ with any seriousness
I should have said I have a few hundred systems out in the field over the last 11 yrs. Basic engineering analysis is based off what worked in real life rather than theoretically should work but I'll add a 50% safety factor just in case (I can't find the :WINK: icon so read it as having been added here :lol

I’m not comfortable with load currents in excess of the individual cell. Cascade failure is a real issue in parallel systems.

In my case each “ battery “ consisting of 4S cells of 100Ah capacity will have battery based current protection.

If that battery current trips the BMS will shut down every battery until manual overrides can be triggered to decide what is appropriate to reconnect. This means system current is limited to individual battery max. I feel strongly this limits cascade failures.

As you have already posted, the LFP cells are reliable, the cheaper end of the scale just has less capacity to handle higher loads because the real capacity is a fudged figure, generally with the "you can only have 80% SOC " so in reality the battery is actually only 80% of what they advertise on the case. The next most common fudged technical bit is the load that capacity was tested at, and this is the true load limit in any single cell ..... advertised 100Ah capacity does not say you can pull 100 amps at any time and unlikely for an hr to get the full 100Ah of advertised capacity.
As an example, often the lower end of the market like to use the more common C20 rate for load testing and draw the battery down until the voltage drops like a stone .... and this is the point they measure the capacity. C20 for a 100Ah battery is only 5 amps load, if it is fudged "80%" figure, then the C20 rate is only 4 amps.
If you want to use the C1 rate as the base for max load, you need to test just how many Ah you get out before the cell drops below 3v while under load, then reset the cell capacity and C1 load from there. In the end, if you do the sums as to $$ per usable Ah, the Winston cells aren't all that expensive.

I commend you for attempting to make the drop in market a bit more reliable, it certainly isn't at the moment ....... but I still make a few $$ pocket money trying to at least get them to work together .... and that is the hard bit, full load is fairly easy, it's that small parasitic load that causes the hassles, a 1 amp load over 24hrs will pull the battery with the least total circuit resistance down to 75% capacity, yet the voltage will remain the same across all the parallel batteries .... you understand accumulative resistance so it doesn't take much to realise why one battery will end up supplying all the light loads for at least 50% of its capacity, yet a balancer that relies on voltage to sense where the capacity needs to be drawn from and where it needs to be sent to will not rebalance the the batteries, only cells in parallel and those groups in series can do that .....

T1 Terry

I agree with you re the cell capacity. I’m looking at 30% headroom over max design requirements. And yes the $$$$ have to stack up. I’m using GWL oem brand batteries.

I also agree re the major issue is the variability of the series battery resistance resulting in one battery doing all the work , even though as the terminal voltage falls , the other batteries will take some load.

I agree that the big parallel strings mitigate against that but I simply don’t like the unmonitored cell issue nor the ability to retain a working battery after cell issues. In my case I have the ability to quickly take a battery off line whilst retaining remaining AH

Your parasitic load issue is interesting , as my BMSs will all have local mosfet disconnects one could postulate a battery strategy where the BMS would selectively disconnect batteries that are over serving the small loads. This would then spread the load around the battery Bank as each BMS is tracking local battery SOC. so this could be used as strategy to help balance discharge rates ( too a point )

[goboatingnow]

Originally Posted by s/v Jedi

Read the original yaBMS thread on why that isn’t a good option and how, in the end, the INA family is the best option. The INA226 has a 16-bit ADC or 0.5mV accuracy. This may double when subtracting the reading from the previous cell, so 1mV. If you need higher precision then there are 20-bit versions, I think the 229.

The debate on the yaBMS on the parasitic draw of resistors is just nonsense really Simple buffering and very high value resistors easily solved that problem.

[s/v Jedi]

Originally Posted by goboatingnow

The debate on the yaBMS on the parasitic draw of resistors is just nonsense really Simple buffering and very high value resistors easily solved that problem.

Gee that we didn’t think of that!

[s/v Jedi]

Originally Posted by flightlead404

This is a great thread and I'm enjoying reading it, thanks Jedi for starting it.

I've got a fair amount of experience with Arduino over the last decade and I'd like to add one thing I've learned. The el cheapo Arduino you get on Amazon or ebay may be fine for experimenting, but when it comes to building something permanent, especially something as important as a BMS, I'd recommend going with the name brand real Arduino.

I've lots of experience with the cheap ones, and I've found they tend to fail. Sometimes physically, eg component soldering or traces, other times for no apparent reason.

Thank you The new MKR series from Arduino is great but they use an ARM Cortex instead of an ESP. I don’t care about which core as long as they have Arduino IDE support and library compatibility. It seems they all have that. I like the Teensy units as well because of the power and features like dual CANbus on board etc.

I have not had a single clone Arduino fail yet, not even when I use EEPROM writes abundantly… trying to make it fail.

You will see many DIY builders debate endlessly over price, especially when it’s about more expensive components like lfp cells, battery disconnects etc. but even for the microcontroller they always like the cheapest ones… which for an expensive battery like lfp house batteries of 10kWh, will be a difference of maybe twenty bucks. They tell me it all adds up but so do the reliability risks

[goboatingnow]

Originally Posted by s/v Jedi

Gee that we didn’t think of that!

Well there’s was about 10 posts arguing current through the dropping resistor was draining a 1000 Ah system.

Basic electronic volt metrology is well established

I’m going to use 1:1 instrumentation amps that have 2x common mode voltage over supply voltage

Adc reference is more or less cell max voltage. Hence 12 bit resolution more then adequate. No need for opto isolated i2C

Simple efficient cheap and functional

[s/v Jedi]

Originally Posted by goboatingnow

Well there’s was about 10 posts arguing current through the dropping resistor was draining a 1000 Ah system.

Basic electronic volt metrology is well established

Gee, we should’ve used higher resistance values

[goboatingnow]

Originally Posted by s/v Jedi

Gee, we should’ve used higher resistance values

And a buffer amp, ie your common or garden DMM

Or differential full scale measurements needing no droppers and providing a gnd referenced output