Solar System

Build thread (german) at https://chaos.social/@patagona/113149877881417810

Overview

This is sort-of build log / lessons learned of my off-grid solar system / battery backup.

Specs/Features:

• ~7.2kWh battery
• >5kW AC out
• 5kW PV in
• 2kW AC in (with configurable power limit)

Batteries

Basically, what drove me to start this project.

Someone was selling three Pylontech battery modules (1x US2000C, 2x US2000, each 15S LiFePo4, 2.4kWh) for really, really cheap (<25€/kWh) because they were water damaged.

So of course, I picked them up and as expected, the BMS PCBs were corroded to the point of not being feasible to repair (mostly due to the half-assed "conformal coating" applied to the PCBs), but the battery cells themselves were fine, albeit still a bit wet, limescale everywhere and a bit unbalanced.

After opening up the packs, letting them dry, cleaning up the limescale/corrosion and balancing them, all of the cells seem to be fine.

BMS

To replace the corroded BMS I decided to get DALY K-Series 15S 100A "Smart" BMS modules.

I've since replaced these BMS modules again with DALY 100Balance BMS with 1A active balancing because the passive balancing of the K-Series is a bit weak for these huge cells.

Unfortunately, when I started this project, they were not available yet.

Mount

To mount the BMS, I designed some adapter / mounting plates that fit the original mounts / standoffs of the Pylontech battery modules.

The design files can be found in cad/battery/bms-mount.

Protocol

The BMS has a UART, RS485 and CAN interface. There is already a DALY BMS integration using a UART interface in ESPHome, however that is for the older DALY BMS, not the new H/K/M/S-Series (for which there is already a GitHub issue).

After writing an entire RS485-based ESPHome integration, I've realized that it's not really usable due to the non-standard "Modbus" implementation and other issues (see BMS Communication).
I've since written a CAN-based integration available as patagonaa/esphome-daly-hkms-bms-can that works better.

Case

For transport, the batteries are mounted in a used flight case. I'm using an Anderson-style connector as interconnect between battery and inverter (and other loads), which I've mounted towards the outside of the case so it can be used with the lids on.

There are two parts two the mount: one that holds the connector (cad/battery/battery-sb120-mount/battery-connector-mount.scad) and one for between mount and plug so the mount doesn't get squeezed when tightening the screws (cad/inverter-box/frontpanel/handle-filler.scad). The mount has later been adapted to be a lot stronger (due to not having to fit in the tight space on the lid) for the front panel (cad/inverter-box/frontpanel/sbmount.scad).

Inverter

I decided to go with a "EASun ISolar-SMH-II-7KW" (which is produced by the ODM "Voltronic", similar/equal devices are also sold under the names "MPPSolar", "PowMr", "Powland", "FSP", ...), mostly because it was cheap (<300€) and has a lot of output power.

Datasheet: https://drive.google.com/file/d/1s9XwkGRsnE1rK3pe3qbGn26EQGxV9woU/view

Power Limit

On battery power alone, with my 15S LiFe battery, the inverter already showed 99% load at around 5.4kW instead of the advertised 6.2kW. Further testing has showed that this is due to a (~120A) current limit of the DC-DC boost converter that generates the internal ~350V DC link from the 48V battery voltage.

When combined with solar input (which also goes to the internal DC link), the inverter has no problem outputting 6.2kW continuously and even handles peaks of 7.2kW before switching off after 10 seconds (as specified in the datasheet).

Protocol

The inverter has a RS232/RS485 interface which can be used to read data from the inverter (using a custom mostly-ASCII-based protocol) and which has been implemented by several people already:

• https://github.com/jblance/mpp-solar
• https://esphome.io/components/pipsolar.html

Inverter Box

The inverter is mounted on a plate in a 60x40x22cm euro container for easy transport and (at least minimal) water-resistance, along with all the other necessary support hardware (circuit breakers, etc.).

Files in cad/inverter-box/box.

Front Panel

There is a laser-cut front panel exposing all the important connections (DC, AC, PV) on the short side of the euro container.

Files in cad/inverter-box/frontpanel (also includes the SB50/SB120 mounts, DIN rail mounts, etc.).

Air Duct

To prevent overheating, an air duct was designed (in Fusion) and 3D printed to direct the warm air from the inverter fans directly out the Eurobox handle.

Files in cad/inverter-box/air-duct.

PSU

To be able to limit the AC power input (for charging the batteries with a small generator) a Huawei R4830 PSU has been added to the inverter box as well.

See my Documentation and ESPHome component for more info about these PSUs.

For this, a second level has been added to the inverter box to hold the PSU, the necessary power distribution blocks and circuit breaker and some additional unrelated hardware (auxilary power circuit breaker, ESP32 board)

Misc other CAD parts

• cad/inverter-box/frontpanel/sbmount.scad: Panel mount for Anderson SB120 and SB50 connectors
• cad/inverter-box/din-rail-plug/: Part that fits into the cutout for the DIN rail parts so not each unit has to be filled
• cad/esp-box: Case that fits a LilyGO TTGO T-CAN485 board (with CAN bus and RS485) and an RJ45 breakout board for the inverter RS485
  • v2 splits ESP and breakout board in separate cases
• cad/battery/battery-fpconnector-bus: Mount for pluggable screw clamps which fits in the Battery RJ45 hole to interface with the BMS RS485
• cad/battery/battery-fpconnector-aux-power: Mount for an Anderson Powerpole connector which fits in the Battery RJ45 hole to power an ESP32
• cad/lxcharger-end-cap: End cap to replace the default barrel jack of an LXCHARGER H65AC DC-DC USB-C charger with an XT30.
• cad/inverter-box/mcb-mount: Part to mount din-rail miniature circuit breakers on a board sideways
• cad/inverter-box/top-holder: Holder to mount a board above the inverter to have make space for the PSU, power distribution blocks, etc. (printed with a pause to insert screw nut).

You Get What You Pay For: RS485 Transceiver Modules

The (pretty common) RS485 transceiver modules I used for inverter communication do have a ground pin on the RS485 side. This does, however not mean it is actually connected to anything meaningful.

Actually, there is no connection between the TTL- and RS485-side ground, which means any stray electric current flows directly into the A/B inputs of the RS485 transceiver.

In some cases (for example, when both the PV input and the MCU are ground-referenced, even just through a Y-capacitor of a PSU) there may be something like 90V AC at 700µA between the two sides of the RS485 transceiver, killing it pretty much instantly.

Running a wire from the TTL-side ground to the RS485-side ground fixes that issue by giving the stray currents a direct path to flow.

You Get What You Pay For: Inverter

Startup Voltage

At one point, the inverter just stopped working (when switching it on, it turned off itself after a few seconds). Turns out, there is an (undocumented) "cold start" voltage, that is hard-coded to be ~4V above the low-voltage cutoff. Means, when the low voltage cutoff is set to 45V, the inverter will just not start below ~49V.

Also, the battery voltage readout was wrong by around 0.7V, however that could be fixed through a serial command (can be sent using mpp-solar: mpp-solar --porttype serial -p COM4 -c BTA+01 -I to increase voltage reading by 50mV, can be run multiple times until the value is correct).

Questionable Isolation, EMF

The different parts of the circuit (PV charger, battery DC-DC, AC Output) are isolated from each other, but still loosely coupled via (stray?) capacitances.

When the AC output is on, it seems like the H-bridge switching the ~350V DC bus during zero-crossing causes a short but large voltage spike (~300v) that is coupled back to the PV input and can be measured between the PV input and PE (and felt on the terminals!).
This means, if PE is grounded, the solar panels and their wires become an antenna that radiates those pulses as EMF (probably, I didn't have anything to measure with).
If PE is not grounded, the solar panels lying on the ground make a great plate capacitor to ground, making the PE and everything hooked up there an antenna instead.

Those issues could probably be fixed / mitigated somewhat with an extra RF filter, however a 32 Amp RF filter sounds expensive.

You Get What You Pay For: DALY BMS

UART BMS Communication

The DALY BMS comes with two TTL UART ports, CAN and RS485.

There is a UART DALY BMS integration in ESPHome, however it turned out that the new (2023?) lineup of BMS from DALY (H/K/M/S series and 100Balance) use another communication protocol. This new protocol is called "Modbus" internally in the PC software and looks just like regular Modbus.

After some messing around with Modbus software and not getting it to work, I figured out that the reply from the BMS has a different device id than the request, which makes it completely incompatible with all available Modbus software / libraries.

Additionally, the requests sometimes timed out, so I implemented the CAN protocol and switched to that instead (see above).

Current measurment

As it turns out, the BMS current measurment is messed up in several ways (see below). This also affects the couloumb counting / state-of-charge calculation.

Miscalibration

At the same current, the three BMS all showed different currents (up to a few 100mA). This could be solved by doing a calibration of all of them at 0 and 10A

Sampling Rate

The current measurement sample rate seems pretty low and there also seems to be no averaging at all. This means, when the AC inverter is on and pulls e.g. 10A RMS (as 50Hz sinusoidal half-waves), when polling the current every second, the readings may be 2 -> 12 -> 5 -> 1, instead of just showing a constant 10A.

High Dead Zone / Low accuracy ADC

The BMS simply does not measure anything below 100mA. When the inverter is off and there is just a small load from the ESPs etc., the BMS will just not measure anything.

You Get What You Pay For: LilyGO T-CAN485

Again, how hard could it be to do RS485? (answer: apparently very hard)

They use an MAX13487E transceiver, which has "AutoDirection Control", which means instead of driving the bus in both directions, resulting in in [0.4V, 5V, -5V] (idle, mark, space - with proper termination and bias), it only drives it in one direction, relying on the bias for the other direction, resulting in [0.4V, 0.4V, -5V]. This isn't great interfereence-wise, but it works and is spec-compliant.

However, LilyGO chose to terminate the bus with 120Ω and bias it with 4.7kΩ, resulting in the voltages [0.06V, 0.06V, -5V], which just barely may or may not work. They could've omitted the termination, done the biasing right or just use a "normal" transceiver (they have enough GPIOs to do so), but it seems they've instead chosen the perfect combination of parts to make it not work 🤦.

Additionally, they also did the same as #1: Transceiver Modules, so the protection circuitry doesn't really protect the transceiver as the protection ground isn't hooked up to the transceiver ground (besides a weak connection through 1MΩ + 1nF).