APC UPS LTO upgrade

Following my work on the decoding of the Microlink serial protocol, I wanted to replace the batteries in my SMC1000i with fresh ones. The original battery pack part number is RBC142 and consists of two 12V 9Ah lead-acid gel blocks, connected in series using 6.3mm spade connectors. They can be replaced by non-original ones quite cheaply. Due to my long-time involvement with lithium batteries at work, I started to look at how feasible it would be to assemble a replacement upgrade-pack instead.

The target

This particular UPS (Smart UPS SMC1000i) has an inverter that can supply up to 600W to the load. At 24V, this means >25A is drawn from the batteries at full load! Discharging a lead-acid cell at almost 3C is pretty extreme. Considering that the capacity of a lead-acid battery is usually specified for a 20h discharge (0.05C), it is unlikely that the batteries will outlive multiple outages without significant capacity loss.

On the other hand, lead-acid batteries are extremely robust and tolerant to abuse such as overcharging. That is why not many UPSs include any form of charge management beside a CC-CV cycle; it would add cost for little gain.

Lithium battery recap

While lithium cells can perform much better, they need to be protected against over- and underdischarge to avoid the risk of fire or explosion. In practice, this means a Battery Management System (BMS) is added to a lithium battery that monitors the voltages of each individual cell and disconnects the pack in case of a potential problem (such as over- or undervoltage and too high or too low temperature). Furthermore, lithium cells, when overcharging, do not 'bleed off' the excess charge in the form of hydrogen as lead-acid cells do. The voltage will keep rising over 4.4V, at which point the electrolyte will break down, heat up and eventually catch fire. Thus, a BMS also needs a way to 'bleed off' the excess charge of each cell, which is typicall achieved by dumping excess charge into a so-called balancing resistor.

One specific type of chemistry is the Lithium-Titanate cell. As opposed to conventional lithium chemistries such as NCA and LFP, they do not use graphite on the anode. This property makes them tolerant to over-discharging and low-temperature use. They also boast very high charge and discharge rates, up to 10C, and a longer life. The downside is their lower voltage, which is usually in the range of 1.8-2.6V, that also limits their energy density. I decided to build my UPS battery from LTO cells to evaluate their useability.

Sourcing small LTO cells locally proved difficult, as only high-capacity cells (40Ah and upward) are available. Eventually I bought twenty 2.5Ah cells from GTK in China. Customs took a long time due to incomplete invoice though.

Balancing/protection circuit

While LTO cells can be safely discharged to 0V, overcharging and imbalance in a pack is still to be avoided. I wanted to avoid adding a full BMS to my pack as that would increase the cost again. Since the voltage is in the same range (2.7-2.8V), I considered the use of ultra-capacitor (also called EDLCs) balancing/bleeder circuits. The maximum balancing current of these is usually very small, and dedicated ultracap balancing ICs proved to be surprisingly expensive (such as the Rohm BD14000EFV).

My search led me to balancing circuits based on Zener diodes. Unfortunately, Zener diodes in this voltage range have a large spread on the trigger voltage. So some cells in a pack could start balancing at 2.7V and some at 2.9 for the same part... An alternative is the use of a shunt regulator, such as the TI TL432. This component basically acts as a relatively accurate and adjustable Zener diode.

The TL432 can only tolerate a modest amount of current, so it has to be coupled to a transistor to be able to balance at least the 100-200mA I am aiming at.

The charging circuit in the APC UPS seems to stop charging at ~27,3-27,4V. For my string of 10 cells, this means they get 2.74V, while their specified charging cut-off voltage is 2.85V. So, I chose the shunt regulator to start at 2.8V. This is chosen by a set of 1% resistors. Once above 2.8V, the regulator will try to maintain that voltage by controlling the current through four parallelled 1Watt 10 Ohm SMD resistors.

PCB design

A circuit was designed in Kicad so that a single PCB could hold both the 2p10p cells and balancing circuits. This pack would give a theoretical 5Ah capacity at 24V. This is less than the original 9Ah, but I still expect a longer runtime from it as the useful capacity during (high) load should be much higher.

I tried to ensure adequate heat dissipation by soldering the MJD45H11 transistors' backside pad and the dissipation resistors to an as large as possible copper area, without thermal reliefs.

To obtain a strong connection for the bolted connections of the cells, a field of filled vias was used. I saw this somewhere and it seemed like a good idea. Eventually, the PCB looked like this.