Power module for a multichannel battery tester

As the “repair” project gradually winds down, I’m slowly finding time to get back to my other projects. I’ve started with a multichannel tester for lithium battery packs, which I first wrote about last year.
There will be several stages, but I started with the most difficult and substantial one: the power module.


As a reminder, a friend asked me to build a device that could charge and discharge lithium battery packs up to 19S. Of course, 20S would have been a nice round number, but this is limited by the maximum voltage of the adjustable power supply, so I had to settle for 19S.
19S is 4.2 × 19 = 79.8 V. The power supply can deliver a maximum of 80 V (link), leaving 0.2 V for voltage drops in the wires and so on.

I think many people (if not everyone) who use large battery packs have either dealt with balancing themselves or at least heard of it.
By large batteries, I mean, for example, packs used in hoverboards. They originally have just two wires coming out, but if you open the battery, you’ll find a connector with all the pack connections brought out—in other words, 11 contacts. This connector lets you connect directly to the cells, bypassing the controller.


For 6S batteries, you can still find a charger with balancing, such as the iMax, but 10S is much more difficult, not to mention 15S and higher, where the end-of-charge voltage exceeds 60 V!
For example, I had to balance the battery shown in the photo below by hand. You can see my notes from the process.


So, what is balancing?
Batteries aren’t perfect to begin with, and they also age at different rates during use. As a result, their capacities and internal resistances can differ significantly. An unbalanced battery pack isn’t dangerous in itself, provided you have a protection board, of course, but you lose capacity: the upper cutoff is determined by the most charged cell, and the lower cutoff by the most discharged one. From my experience testing batteries like these, an unbalanced pack might deliver, say, 1 Ah out of its rated 4 Ah, while after balancing it can deliver 2.5–3 Ah.

Balancing is the process of equalizing the voltage across the cells. For example, you can either fully charge them all or fully discharge them—in other words, bring them to the same state.

There are several ways to do this. The simplest is passive balancing.
This approach is used on the vast majority of protection boards because it’s the easiest to implement.
As a cell reaches its threshold voltage, a resistor is connected across it, diverting some of the charging current. While that cell is “struggling,” the others have time to charge to their maximum.
Here are a few illustrations from this article.

1. One cell is either more charged than the others or has slightly less capacity.
2. During a regular charge, its voltage will be higher than the others.
3. The balancer diverts some of the charging current, preventing the voltage from rising above the maximum.
4. As a result, all the cells are charged evenly.


A balancer like this can be built right into the protection board; it’s highlighted in red here.


They’re also sold separately, but both types have a major drawback: their balancing current is limited to around 50–60 mA, which isn’t enough when the imbalance is significant.


There are also active balancers. I’ve built similar ones for radio stations.


This type of balancer transfers energy from a charged cell to a discharged one. The current is very low, but since the circuit runs around the clock—not just while charging—it still helps.


As an example, here’s my test of an active balancer. Over several days, it equalized the cell voltages.


If this approach isn’t powerful enough for you, high-power balancers are available. They work on the same principle, but with a balancing current of around 1 A. I have one of these boards at home and plan to write a review of it. The downside is the price, unfortunately.


Another option is to charge each cell independently, with no limit on the number of cells. You can use small charger boards. They’re fairly inexpensive, but this approach has its drawbacks:
1. The power supply must have as many galvanically isolated outputs as there are channels to charge.
2. The charging current is fixed and cannot be changed on the fly.


For this device, I wanted to balance the cells quickly, ideally with a flexible operating mode and the ability to set different charging currents.
So I decided to make a kind of hybrid: a balancer and charger. I wanted a charging current of up to 1.5–2 A.

I based it on the principle of a passive balancer, where the circuit tries to limit the battery voltage by shunting it through a 50–100 Ω resistor. But a resistor-based circuit didn't work for me for several reasons:
1. Limited flexibility
2. It's difficult to dissipate heat from the resistors. At 2 A and 19S, that's about 160 W!
3. This may sound strange, but cost is another issue: high-power resistors that can be mounted on a heatsink aren't exactly cheap either.

So I decided to build a shunt voltage regulator based on the well-known TL431 and an additional transistor.
Advantages:
1. Much more flexible: the circuit limits the voltage equally well at both 500 mA and 2 A, and I think the current could be increased further with additional cooling.
2. The transistors are designed to be mounted on a heatsink, so mounting them is straightforward.
3. A transistor costs less than a high-power resistor.
4. Smaller circuit board.

The regulator circuit is fairly well known. An adjustable Zener diode monitors the input voltage. When the voltage rises above the set level, the Zener starts turning on the power transistor, and from there the circuit simply operates in linear mode.
Essentially, it's just an accurate, high-power Zener diode, which is exactly what I need.


In practice, this gives me a multichannel charger. You just need to supply it with a voltage greater than 4.2 V × the number of cells and limit the current to the required level.
As an example, I drew a circuit with 6 channels. The Zener symbols represent the nodes whose circuit I showed above.

1. You can connect all six batteries.
2. You can connect just three in a 3S configuration.
3. Or connect three batteries to any channels you like.

In all three configurations, the batteries will charge at the set current up to 4.2 V.
The downside of this solution is the heat dissipated in unused channels, or in channels where the batteries are already charged, since the Zeners will start absorbing all the excess energy.
4. To solve this problem, the positive output of the current source can be connected to any point in the battery pack. For example, with a 3S battery, you can connect it to the positive terminal of the top cell. In that case, you can set a lower voltage, but since the circuit needs a current source to operate, you can simply leave it as is. Technically, you could simply set the output to 80 V, set the current limit, and use it like that. The only downside is that it will spark when you connect the “floating” terminal.


Next, I’ll describe the assembly steps and the problems I ran into along the way.

To avoid problems with both quality and consistency of the parts, I went to our local Kharkiv Cosmodrome and picked up the components:



20 TL431 zener diodes — link
20 BD438 transistors — link
It came to about 10 UAH ($0.4) per channel, or $8 in total.
I also bought a couple of trimmer resistors.


TL431 from Texas Instruments, 1% accuracy.
I spent a long time choosing the transistors. I needed PNP types with decent gain, convenient mounting, and enough power dissipation. The TO-220 options usually weren’t suitable for this job, so I ended up buying BD438s in the SOT-32 package, familiar to local radio hobbyists from the well-known KT814–817 series.


I put together a prototype using point-to-point wiring for testing. I calculated the resistor values with the program at this link.


It actually worked pretty well right away. Without any adjustment, I got 4.17 V at 0.5–1 A (the two photos above).
But I wanted better results, so I added a trimmer and easily set it to 4.2 V. Doubling the current from 0.5 to 1.0 A raised the voltage by just 1.6 mV!


Unfortunately, this version isn’t suitable as the main balancer because the circuit’s own current consumption is around 2–3 mA.


I also couldn’t get the indicator to work. Adding an LED between the zener diode and the transistor base raised the threshold voltage significantly; the circuit simply didn’t have enough voltage across the zener–LED–base-emitter junction path.


The switching is very precise. For example, going from almost zero current to 100 mA requires increasing the input voltage by just 10 mV. Further increases in voltage raise the current to the level set on the power supply, which then switches to CC mode.
Overall, I was happy with this part.


This is the final circuit. The divider resistors are critical both in value and accuracy. The 6.19 Ω resistor was an accidental find: the store didn’t have anything closer, and there was only a tiny price difference between 1% and 5% parts.


Then it was the usual process: I laid out the PCB, transferred it to the copper-clad board, etched it, and tinned it.
While I was working on the board, a friend bought the remaining parts: resistors, a thermal relay, and so on.


Choosing the resistor values took a long time. I wanted the trimmer to have the lowest possible resistance while still providing a reasonably narrow adjustment range.
I also bought precision resistors, mostly 1%, with a couple of 5% parts whose resistance doesn’t matter much. The 1 MΩ resistors were bought in case I needed to make fine adjustments.


Let’s assemble one channel, apply power, and see what we get. I calibrated it at 1 A.
1. 0.5 A
2. 1.0 A
3. 1.5 A

Given that the maximum allowed is 2.45 V, I think the result is pretty good.


As you saw above, I also bought 4.99 kΩ resistors. The idea was to do without the trimmer: in that case, a 4.99 kΩ resistor replaces the 5.1 kΩ one, and the trimmer’s contacts are shorted together. The example below shows the second channel.


It works pretty well this way too. At 0.5 A, I got exactly 4.2 V. Here are the results at 1 and 1.5 A.


The current-versus-voltage response is excellent: on the left, the current in the circuit is almost zero; on the right, it’s already around 40–50 mA. The voltage difference is just 6 mV.

But after thinking it over, I decided against this option because I still wanted to keep the adjustment feature.



Since this section can dissipate up to 85 W at a charging current of 2 A, proper cooling is essential. For that, I bought a large heatsink and a protective grille from the same Cosmodrome. I found a fan at home.


The heatsink has six fins, each with fine grooves to improve heat dissipation.


The 300 mm heatsink weighs almost a kilogram. This is one of those cases where buying locally is cheaper than ordering from China: mine cost $8.50.


1. A little shorter than the stated 300 mm
2. Almost 35 mm tall
3, 4. The body is 7 mm thick in the middle and 4.5 mm near the edge.
The heatsink surface is smooth and flat, with no dents.


It's about 83 mm wide. I took this photo separately to show the spacing between the fins.


Next, as they say: “Saw away, Shura, saw away.” It took a long time, tedious work with an ordinary hacksaw. Unfortunately, this was the only suitable option in the store. There was also a 100 mm one, but it was too short for me, and they didn't have a 150 mm one.


Now we swap the hacksaw for a drill, bits and taps, and keep making aluminum shavings.
I ended up with about 60 holes in total, most of them threaded:
20 for the balancing board transistors
6 for mounting the balancing board
4 for the load transistors
4 for mounting the load board
8 for the fan and grille
8 to connect the heatsinks to each other
4 for the thermal switch
8 for the wires to the thermal switch (4 turned out to be unnecessary)

There are 4 more holes for mounting the heatsink itself, but I'll drill those once it's in place.

The design is simple and straightforward. I tested it successfully in my previous electronic load, but that one had a worse heatsink, one and a half times shorter.
While I was at it, I test-fitted the electronic load board and the fan. The gap between the heatsinks is 5 mm, and the 80 mm fan mounts in holes in the heatsink base.


I bought IRF640 transistors for the load — link — and normally open thermal switches rated for 55 degrees.


I usually prefer insulating the heatsink from the case rather than insulating the transistors from the heatsink, but that wouldn't work here. There's no way I could cut the heatsink into 20 pieces and then insulate them all.
So I took a strip of mica, split it into thin layers, applied thermal paste and loosely fitted the transistors for one board.


Just to be safe, I checked that the transistors weren't making contact with the heatsink.


I fitted the board onto the transistor leads. This was the most awkward part, and I still haven't figured out a more elegant way to do it, unfortunately...


Then I set the gap between the board and the heatsink to 5 mm, soldered one lead on each transistor, and carefully removed the board with the transistors attached. Then I soldered the remaining leads.


While screwing and unscrewing everything, and drilling the mounting holes in between, I realized I'd been doing it the wrong way. It's easier to first screw the transistors in place without the mica or thermal paste, then solder them, drill the board mounting holes, and only then fit the mica and apply the paste. So that's what I did with the second board.


First power-up: no smoke, no short circuit, and with almost 42 V at the power supply output, I got a current of 140 mA.


I raised the voltage, and the power supply went into CC mode. The voltage across the 10 channels was 42.057 V, or 4.2057 V per channel.
Then I tried to calibrate it. I used the lower board contact (ground) as the reference point, then adjusted the channels to read 4.2, 8.4, 12.6, 16.8, and so on. After calibration, the total came to 42.015 V. The overall voltage wasn't really tied to the actual state of things: I raised it on some channels and lowered it on others. it's clearly an "average temperature across the whole hospital".


Along the way, I ran into a problem: I couldn't calibrate one channel, and the lowest voltage I could set was 4.2086 V.
I tried replacing both divider resistors, but that didn't help. In the end, I soldered another 1 MΩ resistor in parallel with the upper resistor, as I mentioned earlier.


The second channel calibrated perfectly. The voltage before calibration is on the left, and after calibration on the right. In the end, I calibrated each channel individually, rather than relative to the negative wire as I did the first time.


While I was calibrating, the heatsink had time to warm up quite a bit, even though the fan was off.


But there were still a few "quirks". While calibrating, I noticed that one channel's voltage was too high, but I was able to bring it back to normal without any trouble. Later, I found out that one of the trimmer resistors was 500 Ω instead of 100 Ω like the others.


Since the heatsink is going to dissipate heat from the load as well as the balancers, I decided to make a few improvements to the load too.
I wrote about the load earlier in another review.


Changes made:
1. Replaced the 0.22 Ω shunts with 0.5 Ω shunts (two 1 Ω resistors in parallel).
2. Replaced the 1 nF compensation capacitors with 22 nF ones. With low-capacitance capacitors, parasitic oscillation can occur under certain conditions and with some power supplies, showing up as increased ripple. Capacitors in the 22–100 nF range should work.
3. Added a 120 Ω resistor in series with the fan's power supply (two 240 Ω resistors in parallel) so it would keep spinning at low speed. The thing is, even when the heatsink is cold, there are components inside the enclosure that get warm, and that heat needs to be drawn out.
I considered adding a capacitor in parallel with the resistors to help the fan start up, but for now I've changed my mind, since that would increase the load on the thermal switches. I'd also need to add resistors of around 1–2 Ω.

Before the modifications on the left, after on the right.



And here's the finished module. I chose this heat-dissipation setup because only one thing runs at a time: either the load or the balancer. In both cases, the power being dissipated is about the same, 150–170 W.


The thermal switches are connected in parallel. I installed two only because of the balancers, since unlike the load, they can heat up independently of each other.
The load transistors are tilted because I miscalculated where to drill the hole for the wires to the thermal switches.


Top view.
By the way, I'll have to replace the multiturn potentiometer too, since I don't need such fine adjustment. A stepped adjustment in 0.5 A increments would have been more than enough.


A few quick tests.
First, let's connect the load. Changing the shunt values worked out really well: with the most common 10S batteries, the power dissipation is right around 150 W. So even by pure chance, you can't burn out the load.


The temperature, however, is another story. After a while, some of the transistors heat up to over 100 °C, which is really bad considering they're operating in linear mode.
Interestingly, in the thermal image of the load board, the shunts on the right are cooler than the others, even though the voltage across them is the same.


But after letting it heat up for a while longer and taking a closer look, I found that the actual temperature was closer to 115–125 °C.
For now, I'm blaming the small TO-220 package, which can't dissipate all the heat quickly enough because of the mica insulator. Also, my load has two fans, one of which has a high airflow rating. This one uses a standard computer case fan.
For now, I can see two possible solutions:
1. Replace the transistors with something in a TO-247 package, such as the IRFP460.
2. Install a higher-performance fan.

There are other possible solutions, but I'm not considering them because they're too complicated or not effective enough.



Now let's connect one half of the balancer. For the test, I set the current to 2 A, the maximum I plan to use. In practice, the current will more likely be in the 0.5–1.5 A range. The module dissipates almost 85 W.


Things are much better here. When the thermal relay switches on, the transistor temperature is 72 degrees, then it rises to 77 and barely increases after that. Since these are bipolar transistors, this is a perfectly normal operating temperature, with plenty of headroom.


So, the most complex section is almost ready. There are a few minor issues, but I’ve drawn my conclusions, and the problem should be fixed soon. I’ll report back on it in the near future.
For now, the biggest challenge is heat. Everything else works perfectly. I especially liked how the balancer works—or the multichannel charger, depending on what you call it.

As always, I’ll be glad to hear your questions and suggestions.