Earth Notes: Low-Voltage Disconnect for 12V RE System (2007)
In order to power my low-power laptop/server from my 12V solar PV Renewable Energy (RE) system shared with my 12V office lighting, I need a low-voltage disconnect (LVD) circuit to:
- Protect the battery against damagingly-deep discharge.
- Reserve plenty of power for lighting in the evening.
The laptop actually requires 20V @ 3.25A max, and therefore I need a small and efficient DC-to-DC converter. A car laptop power supply is ideal, and should also filter out nasty spikes, etc.
Note that this is part of a design to automatically switch the laptop back to mains power when there is not enough solar energy to spare.
This design uses a MAX8212 IC intended for battery monitoring, containing a precision voltage reference, and with very low current draw, that needs only two or three resistors to set up.
I use the MAX8212 output to directly drive a high-side p-channel power MOSFET to connect the power to the load (and disconnect when the battery is too low). Since this is either fully-on or fully-off, it dissipates almost no heat.
This also drives an opto-isolator and a LED indicator for visual and computer monitoring of the connect/disconnect state.
I use transient-suppressor diodes and a decent-sized capacitor to help absorb spikes (etc) from other equipment and the wiring runs, and our load should it have any significant inductive component, especially given the fast switching for load connect/disconnect.
This has to be able to handle a load of several amps peak (the peak laptop power-draw is ~70W which implies ~6A continuous). This has to handle a much higher start-up (inrush) current at turn-on.
This must use absolutely-minimal power when the battery is low, since any significant continued discharge may damage the battery (through "sulphation"/"sulfation"). This design draws microamps when it has disconnected the load.
When the load is connected this draws a few milliamps itself.
This uses hysteresis to:
- Avoid oscillating at power-off when the battery voltage rises after load removal. Trial-and-error suggested that this had to be at least 500mV (0.5V), and 1V seems a lot more stable.
- Give the battery chance to get a good charge (ie recover) after a deep discharge.
I have the lower (disconnect) threshold set at about 12.15V and the upper threshold set at about 13.45V. The high threshold should generally ensure that the load is reconnected only when the battery is being charged in good sunlight. (In fact, spikes from other equipment can cause this to be exceeded log enough to trigger the LVD to reconnect the load briefly.)
Given how complicated the relationship of State-of-Charge (SOC) of the battery to voltage is, the threshold and hysteresis values are difficult to get right, especially as I want to interact well with the SHS-6 LVD for my lighting. However, it is clear that there is about 1V difference for a given SOC between the battery being charged and being discharged, so the hysteresis probably has more than this, ie over 1V, which is what I now have.
I have added a small 'smoother' to the voltage-divider network for the MAX8212 to reduce oscillation/instability, especially given the huge start-up current of the DC-DC converter load.
Revision 3: Working LVD
While not perfect, the LVD and 20V switchover parts seem to be reliable, and I feel confident to leave them on 24x7. I can monitor remotely when solar power is being supplied to the laptop, which adds to my comfort.
The MAX8212 and the resistor network R2/R3/R4 sense the battery voltage, turning off power to the DC-DC converter when the voltage gets too low, ie when the battery is getting too discharged, and turning the power back on when the battery has recovered enough and the battery is being charged (the higher threshold voltage is higher than can be seen on a resting or discharging 12V lead-acid battery). The R1/C2 circuit attempts to make the sensing circuit less prone to spikes and to oscillation when starting to power a heavy load. The main disadvantage of this tweak is the possibility of triggering something like 'SCR latchup' on the HYST pin; R1 is there to limit peak current flow and mitigate the latchup issue.
D1 and D2 are there to absorb inductive and other spikes at the input and output and should always keep the voltages at IC1 (and T1) within absolute maximum ratings. C1 is also there to help absorb inductive spikes, and to help supply inrush/start-up current to the load. A larger value (or even a small ultracapacitor) might be a useful upgrade. The power MOSFET T1 is hugely overrated for the typical 2A to 6A at 12V it is expected to handle to help survive spikes etc. Because T1 should basically always be hard on or hard off it doesn't seem to need a heatsink and has never been even warm.
IC1's "OUT" is an open-drain pull-down, active when the battery voltage is OK, and so R6 (and R5/LED1/etc) pull it up relatively quickly to disconnect the load when OUT is inactive.
Basically the relay RL1 is arranged to switch laptop power to be from the DC-DC converter when the converter is running, in turn when the LVD has connected the converter to the 12V battery because the battery voltage and thus charge are high enough. The circuitry around the coil is the help avoid the relay switching too rapidly even if the 20V converter output is not stable for whatever reason (eg oscillation in the LVD). LED2 also drops the voltage available to the coil a little so that the coil's guaranteed pull-in at about 70% of its 24V nominal won't happen until the converter's output voltage is reasonably close to 20V, and thus good for the laptop.
Visual indication is provided by LED1 and LED2 of battery state and the DC-DC converter output actually being adequate to drive the laptop. The optoisolators OPT1 and OPT2 (with the transistor outputs connected to digital inputs on the k8055 USB I/O board) provide exactly the same information to the laptop so that it can monitor where its external power is coming from, since 'external' no longer simply means AC mains. Note that the current for the LEDs and optoisolators is almost 'free' in this design since we have to have for T1 gate pull-up and the RL1 coil anyway. (Possibly the series resistor R8 could be a little higher to reduce power wastage.)
All of this was built on a short piece of copper strip-board, with power connections being handled via terminal blocks and 2.1mm and 2.5mm DC power plugs and sockets, and a car cigarette socket from the LVD output for the DC-DC converter to plug into.
As of ~2008/06/25 the laptop battery seems to have died. This means that the laptop would crash if there was even the slightest interruption in its power supply. Thus the relay-based switchover system was no longer viable. So I established with a digital voltmeter the open-circuit output voltage (both nominally 20V) of the laptop adaptor (20.85V) and the converter (20.47V). I have taken the decision that the DC-DC converter is expendible if need be (and indeed I have a spare), so I have commoned the adpator and converter -ve/0V outputs as before (and they run to the laptop -ve/0V power input), and connected the adaptor +ve output via a P600A 6V silicon power rectifier to the +ve output of the converter and on to the laptop power +ve input. If the converter output is on (LVD output is on) then it will power the laptop since the mains adaptor voltage after the ~0.7V rectifier drop is lower. (Note that a Schottky diode would not have a big enough voltage drop to make this scheme work.) But the moment the DC-DC converter output drops, eg because the off-grid battery voltage is low, then the laptop adaptor should take over instantly without a gap, ensuring that the laptop does not crash. The upside is instant/slient switch-over and no mechanical relay to wear out. The two small downsides are:
- a slight risk to the DC-DC converter in having its output driven to 20V when its input is not powered,
- power loss in the rectifier when the system is on mains power, though that should be much less than 1W in practice.
I have verified that with this circuit power is not drawn from the adaptor when the DC-DC converter has input power, ie the mains adaptor drops back to its typical 2W 'idle' power draw.
As of 2008/07/21 this arrangement seems to be working without a hitch.
High State-Of-Charge (SOC) Detector
As of 2007/09/04 I have constructed a high-SOC detector that feeds into the USB digital inputs that is independent of the LVD circuit. This means that the laptop can tell when the battery voltage (and charge) is high and and the battery is charging, so that (when the laptop is also being powered from the 12V battery) the laptop can use more power to run faster and do important but non-essential work such as disc checking and my AI computations.
.ps XCircuit file.
Note that this effectively an automatic dump load that does computational work rather than running a heating element (the CPU is the heater here!). If you are running something such as a wind turbine which cannot safely simply disconnect when the battery is full (unlike solar PV), then you should still have a normal dump load, but this can help make slightly more sophisticated use of some of the available extra energy if you make sure that the normal dump voltage (typically a little over 14V) is set higher than this monitor/detector comes on at.
The thresholds are set at 13V/13.5V. Much less hysteresis is needed than for the LVD since this does not directly cause an immediate huge battery drain, and any response to changes is mediated in software and may take up to several minutes, so damaging/fast oscillation is unlikely.
I do see induced noise from my WiFi router trigger this, plus what appear to be spikes from (for example) my small turbine, so this can be showing 'high voltage' then the main LVD stays off. The software only takes any notice of this signal when the LVD is on too.
NiMH Battery Charger as Dump Load
As of 2010/07/17 I have bought a Maplin N42FK AA/AAA NiMH battery fast charger that can run in a car (at 12V) and that has a timer and -deltaV cutoffs, ie should be safe to leave completely unattended with cells in. It has a charge-state indicator which should help us know when the NiMH cells are full so that we can take them out and put them in out 'charged' box and rotate others in from out 'uncharged' set.
I have put this downstream of the LVD which was no longer being used for the laptop now that I am using the SheevaPlug for the Internet server. The idea is that when the off-grid SLA battery is full-ish (and charging) then excess energy will be diverted to charging any NiMH cells in the charger. (Also the server will ramp up its work/consumption at its own chosen voltage thresholds as additional 'dump load'.)
This means that when there is excess off-grid solar energy available this can use a little of it recharging NiMH cells semi-automatically. When there are no batteries/cells waiting to be charged then the server gets to use all the energy that it can. In any case this wastes less of what is available and 'free'!
I may want to increase the LVD lower threshold to make this more like a dump, but given that this is a fast charger rather than trickle, possibly the current threshold does allow us to really charge on demand providing that the off-grid system has plenty of energy to spare.