Smart Battery Monitor?

[Toaster]

Just thought I'd make a few comments about the idea of measuring state of charge with only voltage measurements and not current.

On their web site, the SmartGauge people say they use "computer modelling" for calculating SOC. I've done a lot of computer modelling (professionally) so here are some observations. Disclaimer: I am a physicist.

Let's assume the SOC is some unknown function of voltage V, current I and time t. Also throw in battery parameters such as capacity and age, maybe even unknown things. We'll call those BP. Let's assume BP is a slowly varying function of time (they age slowly). And battery temperature is a factor too (also slowly varying).

So, SOC(t) = F(V, I, BP, T, t).

F is a gnarly function, most likely you could not write it down. But it exists.

First observation. The current I is likely to be a function of dV/dt (sorry, this is calculus). Q=CV so I = dQ/dt = dV/dt.C. Of course C may be a function of time but let's assume a slowly varying one (so that dC/dt is negligible). In plain speak, this means we have just eliminated the current I and no longer need to measure it. Magic!!

Then:
SOC(t) = F(V, dV/dt, BP, T, t)

Now, any well behaved function of t can be expanded like this:

SOC(t) = F(t) = A + B.V(t) + C.dV(t)/dt + D.d2V(t)/dt2 + etc., where A, B, C, D, etc are constants. This is just math.

In any computer model, aka simulation, that is based on time, you have to convert continuum physics into discrete computer steps. The longer the time steps (intervals) the more error creeps in. That's likely why SmartGauge measures V at a high frequency - they want to measure dV/dt. In fact, it is highly likely that their model also uses d2V/dt2 and higher derivatives. The higher the derivative, the faster they need the V data.

It's then a relatively simple task for a computer to take a time series of V(t), fit it to the equation and obtain the constants A, B, C, ..... Once you know those constants, the problem is solved, you can convert V(t) into SOC(t). Put in plain English, you can accurately calculate the SOC at time t, if you know the *history* of V.

An important point is that you don't even need to understand the physics of batteries to do this - you can do it by brute force using computers. All you have to believe is that batteries obey reasonable laws of physics.

Give the little chip in the SmartGauge the right algorithm and a starting point, feed it some voltage data, and Bob's your uncle. My guess is it will make measurements, make a prediction for 30 seconds, check the prediction against reality, adjust the parameters, repeat ad nauseum, "learning" as it goes. This is probably how it adapts to battery aging.

It's interesting that they have not yet solved the charging side of things. I suspect that is what SmartGauge V2 will do. There must be some physics making this hard, or perhaps they just cannot acquire V data fast enough, or....

[Dockhead]

Originally Posted by Toaster

Just thought I'd make a few comments about the idea of measuring state of charge with only voltage measurements and not current.

On their web site, the SmartGauge people say they use "computer modelling" for calculating SOC. I've done a lot of computer modelling (professionally) so here are some observations. Disclaimer: I am a physicist.

Let's assume the SOC is some unknown function of voltage V, current I and time t. Also throw in battery parameters such as capacity and age, maybe even unknown things. We'll call those BP. Let's assume BP is a slowly varying function of time (they age slowly). And battery temperature is a factor too (also slowly varying).

So, SOC(t) = F(V, I, BP, T, t).

F is a gnarly function, most likely you could not write it down. But it exists.

First observation. The current I is likely to be a function of dV/dt (sorry, this is calculus). Q=CV so I = dQ/dt = dV/dt.C. Of course C may be a function of time but let's assume a slowly varying one (so that dC/dt is negligible). In plain speak, this means we have just eliminated the current I and no longer need to measure it. Magic!!

Then:
SOC(t) = F(V, dV/dt, BP, T, t)

Now, any well behaved function of t can be expanded like this:

SOC(t) = F(t) = A + B.V(t) + C.dV(t)/dt + D.d2V(t)/dt2 + etc., where A, B, C, D, etc are constants. This is just math.

In any computer model, aka simulation, that is based on time, you have to convert continuum physics into discrete computer steps. The longer the time steps (intervals) the more error creeps in. That's likely why SmartGauge measures V at a high frequency - they want to measure dV/dt. In fact, it is highly likely that their model also uses d2V/dt2 and higher derivatives. The higher the derivative, the faster they need the V data.

It's then a relatively simple task for a computer to take a time series of V(t), fit it to the equation and obtain the constants A, B, C, ..... Once you know those constants, the problem is solved, you can convert V(t) into SOC(t). Put in plain English, you can accurately calculate the SOC at time t, if you know the *history* of V.

An important point is that you don't even need to understand the physics of batteries to do this - you can do it by brute force using computers. All you have to believe is that batteries obey reasonable laws of physics.

Give the little chip in the SmartGauge the right algorithm and a starting point, feed it some voltage data, and Bob's your uncle. My guess is it will make measurements, make a prediction for 30 seconds, check the prediction against reality, adjust the parameters, repeat ad nauseum, "learning" as it goes. This is probably how it adapts to battery aging.

It's interesting that they have not yet solved the charging side of things. I suspect that is what SmartGauge V2 will do. There must be some physics making this hard, or perhaps they just cannot acquire V data fast enough, or....

This is very interesting. I guess that although not all of us has access to the toolbox, all of us grasp the essence of this. It just means that if you can figure out the gnarly function "F", you can invent an algorithm which will derive SOC from V and some other Stuff over Time.


I am no physicist (as my brother, a physics professor, never tires of pointing out), but it seems to me that we can't know a priori whether the gnarly function F even exists, without some understanding of the relationship between V and the other Stuff we have access to, and SOC. I doubt, for example, that there is any F which will give us SOC out of MP and PC, where MP is the current moon phase and PC is the current price of cabbages.

We know that at any given SOC, V will vary according to C, T, and some other Stuff. So getting SOC from V alone is not going to be simple, and I don't think we can know for sure whether it is even possible, a priori. And that is the whole rub here.


SmartGauge does not know C, and does not even know T. C is varying rapidly, and T slowly. I guess that over time it is possible to recognize patterns in the voltage curves which allow you to partially dispense with C and T. Perhaps with time and learning, the program starts to recognize individual loads being put on an taken off, and is able to see SOC in the voltage response to those loads. It is not obvious, however, how this works, or how well it works.

In my particular case, the SmartGauge performs the same (i.e. superbly) as voltage measurement with loads of less than 0.02C [here "C" is "capacity", according to battery industry convention]. If you have to measure voltage with bigger loads on than that, then this doesn't work as well, and perhaps the SmartGauge shows its "smarts" more.


Concerning charging: I think it's quite elementary why this is difficult without knowing C. Voltage has nothing whatsoever to do with SOC during the constant voltage phase of the charging cycle. I don't really see how they will ever deal with this, without adding a shunt.

We can imagine (just), that some very slight and rapid variations in voltage could betray something about SOC achieved during constant voltage charging, since "constant" of course is never absolute.


But there are limits to what is practically feasible in this attempt to know things with very thin data, notwithstanding the power of computers, even if they are using AI and have a chance to learn.

A very dear friend of mine spent several years of his life trying to devise a gnarly function F to predict future short-term movements of the price of various commodities, based on past short term movements. His theory was that there are patterns which can be recognized by applying enough computer power. The famous "dead cat bounce", is an excellent example of this. He had some great achievement in using artificial intelligence for commercial purposes; he invented for example the first voice recognition software for call centers ("Press or say three for English"), back in the '80's, and made his first fortune off that.

He had a team of 20 or 30 experts working on this, for several years, using a supercomputer. He burned through his first fortune plus a bunch of money he raised from outside investors. But it never worked. The gnarly function F was just too gnarly. I suspect that getting SOC out of V during constant voltage charging would be even harder.

[Viking Sailor]

DH,

It is all about curve fitting. Each battery has a unique voltage curve when being discharged. These curves are the result of the design and materials used in the construction of the battery. The operating temperature and the age of the battery effect the amount of energy the battery can store, but not the shape of the curves. Changes in the discharge rate serve to make the voltage measurements noisy. One of the positive affects of curve fitting is that it reduces the effect of errors induced by noisy measurements. Thus, the a battery SOC can be estimated solely by determining the appropriate shape of its curve and measuring its voltage.

The battery also has a unique amperage curve when being charged. However, the battery charging current is provided by a charger with its own unique setup and charge characteristics. This can greatly increase the variation and complexity of the charging curves. Fortunately, the charge curve near 100% charge becomes very flat with a small charging current required. This characteristic is used by modern chargers to terminate charging after a time delay. This does not tell you how much energy has been put into the battery, it only tells you that most of the usable energy as been returned to the battery. Thus, battery system are designed so the charger puts energy into the battery bank, and the monitor lets you know how much of that energy has been used.

With the SmartGauge, even if charging is terminated early, it is reasonably fast at figuring the bank's SOC. Thus, even if the charge and discharge monitoring functions where combined into a single instrument there would still be a delay before an reasonable estimate of SOC would be available. It is this disconnect between these two functions that would make it nearly impossible to implement an instrument with a consistent accuracy specification.

https://en.wikipedia.org/wiki/Curve_fitting

Paul

[Toaster]

Originally Posted by Dockhead

... but it seems to me that we can't know a priori whether the gnarly function F even exists, without some understanding of the relationship between V and the other Stuff we have access to, and SOC. I doubt, for example, that there is any F which will give us SOC out of MP and PC, where MP is the current moon phase and PC is the current price of cabbages.

Sadly, the price of cabbages is not covered under the standard model in physics, but if it were, then I bet F would include it! It's not necessary to know if such a function exists a priori - you just have to try. Like your friend, it doesn't always work - because either it's not possible, or you're not being smart enough (ie nature is smarter than we are ).

Not measuring T seems like another gap in their method. C probably does not vary rapidly on the timescale of the model adaptation.

Originally Posted by Dockhead

In my particular case, the SmartGauge performs the same (i.e. superbly) as voltage measurement with loads of less than 0.02C [here "C" is "capacity", according to battery industry convention]. If you have to measure voltage with bigger loads on than that, then this doesn't work as well, and perhaps the SmartGauge shows its "smarts" more.

I didn't know this, it's surprising. So you are saying if you have say C = 400 Ahr, and your usage exceeds 8 amps (only 100W) then it does not work as well? That's too bad. I can easily imaging my loads exceeding this quite often.

Originally Posted by Dockhead

Concerning charging: I think it's quite elementary why this is difficult without knowing C. Voltage has nothing whatsoever to do with SOC during the constant voltage phase of the charging cycle. I don't really see how they will ever deal with this, without adding a shunt.

If you know the battery is sitting at V, where V is a charging voltage, say 14.0V, and you know that it has sat like that for X minutes, then you can estimate (based on your model) the charge that has gone into it, and hence the SOC. But, you are exactly right that this is where the system will be weakest - it has little "new" data to help it know for sure what is going on. A shunt would help a lot, but one of their marketing strategies is to emphasize how easy it is to install. Just 2 or 3 wires. I'm surprised they don't use a clamp-on current sensor (no shunt needed). Especially for dealing with high current loads/charging where the model is more likely to be weak.

[sailorboy1]

Originally Posted by a64pilot

I don't think you will kill a bank by leaving it in absorption for a few extra hours once, but I would check water level after doing that

Since heading south is mostly an exercise in motoring I have probably ran my new batteries the past month at 5 hours of absorption/day. At the end of the month only 1 battery took any water and I bet that one was like it was when installed.

[boatbod]

Originally Posted by sailorboy1

Since heading south is mostly an exercise in motoring I have probably ran my new batteries the past month at 5 hours of absorption/day. At the end of the month only 1 battery took any water and I bet that one was like it was when installed.

Well that rather depends on your absorption voltage setting. If you charge at 14.8v you'll need more water than at 14.2v. Personally I like to see a little harder charging as it combats sulfate buildup.

[a64pilot]

Originally Posted by sailorboy1

Since heading south is mostly an exercise in motoring I have probably ran my new batteries the past month at 5 hours of absorption/day. At the end of the month only 1 battery took any water and I bet that one was like it was when installed.

Actually based on what I have read from Main Sail, once we cast off its very likely I will set my Outback 80 to stay in Absorption and never let it drop into float, of course I will monitor acceptance rate to see if this is viable or not, but I bet he is right.
While it's not rocket surgery I don't believe maintaining a bank is a set it and forget it either, unless your tied to shorepower like I am now. I believe though that all it will take is an occasional glance at acceptance rate.
Curiously my SmartGauge seems to do a decent job in charge mode, now it doesn't change as it should during charge, it seems to hang at one number for long periods, but somehow about the time I am fully charged, it has caught up and is indicating 100% charge

[Stu Jackson]

Originally Posted by a64pilot

Actually based on what I have read from Main Sail, once we cast off its very likely I will set my Outback 80 to stay in Absorption and never let it drop into float, of course I will monitor acceptance rate to see if this is viable or not, but I bet he is right.

It is so refreshing to see that folks have been reading, and understanding. It's so important for the health of your electrical system. Like Maine Sail says: batteries don't ie, they are murdered.

Well done, a64!

[daletournier]

Originally Posted by a64pilot

Actually based on what I have read from Main Sail, once we cast off its very likely I will set my Outback 80 to stay in Absorption and never let it drop into float, of course I will monitor acceptance rate to see if this is viable or not, but I bet he is right.
While it's not rocket surgery I don't believe maintaining a bank is a set it and forget it either, unless your tied to shorepower like I am now. I believe though that all it will take is an occasional glance at acceptance rate.
Curiously my SmartGauge seems to do a decent job in charge mode, now it doesn't change as it should during charge, it seems to hang at one number for long periods, but somehow about the time I am fully charged, it has caught up and is indicating 100% charge

A64, I may of missed it, what sort of batteries do you have?
On my last boat I had 6v deep cycles and did set the morningstar so it stayed in absorb all day, no float,worked well. Obviously I was able to check water levels, they really only needed topping up every three months or so. Im fan of the 6v deep cycles, I have no hesitation keeping the voltage right upto them.
My new boat has sealed lead acid deep cycles, im still determining best cause of action with these ones. Just added 2x bluesky mppt regulators to replace the power tec reg, big improvement straight of the bat.

Sent from my vivo Y35 using Cruisers Sailing Forum mobile app

[sailorboy1]

What I love about forums is when you post a real life experience and others try to say how it isn't so.

[a64pilot]

I have AGM Lifelines. I'd rather have good 6V golf cart, but they will not fit in my battery box, too tall.
My hope is my current bank will last until Lithium become more mature and mainstream and a little less of a Science experiment and I will go that route.

[lateral]

Originally Posted by a64pilot

Actually based on what I have read from Main Sail, once we cast off its very likely I will set my Outback 80 to stay in Absorption and never let it drop into float, of course I will monitor acceptance rate to see if this is viable or not, but I bet he is right.
While it's not rocket surgery I don't believe maintaining a bank is a set it and forget it either, unless your tied to shorepower like I am now. I believe though that all it will take is an occasional glance at acceptance rate.
Curiously my SmartGauge seems to do a decent job in charge mode, now it doesn't change as it should during charge, it seems to hang at one number for long periods, but somehow about the time I am fully charged, it has caught up and is indicating 100% charge

Yup, rocket surgery and brain science it is not.
Just don't fall asleep on the job.

[zboss]

Originally Posted by boatbod

Well that rather depends on your absorption voltage setting. If you charge at 14.8v you'll need more water than at 14.2v. Personally I like to see a little harder charging as it combats sulfate buildup.

Apparently so does Trojan in regards to their flooded batteries - they recently slightly increased their float voltage (13.2>13.5) and equalization voltage (15.5>16.2).

I have come to the realization that although the preeminence of real voltages determine the longevity of your batteries that adapting a standardized and rigorous routine will do more overall for your sanity. Which is really the goal of our cruising. I've started caring less about actual Ah consumed (our utilization is under control) and started caring more about managing the system from relative values.