onobeka

I believe that's an example of integral windup. When a control system is at maximum output for a while, if nothing is done to prevent it, the Integral control term (the "I" in PI or PID control) increases without limit, and when circumstances change, the control system takes a long time to reduce the large integral term to a reasonable value. The result is typically overshoot, but in this case it's undershoot.
It's a real shame as control systems are a huge part of the inverter firmware's job. Many things need controlling: solar and utility charge current, bus and AC output voltage, DC->AC converter amplitude and phase, and so on. 

Unfortunately, almost certainly not. Voltronic firmware actually checks for the specific values, and NAKs the command if it's not a magic value.
If the inverters happen to be the old 145 V max ones, including Axpert Kings (NOT King IIs), then you could use fully patched firmware with Dynamic Charge Control, which has this exact feature, and in addition does not wear the EEPROM. However, you do need to send a command every so often (one a minute from poor memory), but you could likely work around  that.
Oh wait, it's an iGrid; it isn't even an Axpert, it's an Infini or similar. So that fully patched firmware is not an option. I'm only 90% sure that the Infini firmware also checks for the specific values.

Alright, the battery was replaced. I've charged the first module to 100% (as the others also were), to get to the same voltage and I will set them in parallel. There's almost no difference in between cell voltages of the new module (0.014V) at 100% and no temperature increase.
 
Thank you very much for all your support!

Alright, the battery was replaced. I've charged the first module to 100% (as the others also were), to get to the same voltage and I will set them in parallel. There's almost no difference in between cell voltages of the new module (0.014V) at 100% and no temperature increase.
 
Thank you very much for all your support!

We are making progress!!!
This is a wealth of info to use to get better acquainted with your battery...
I think there is no need to jump to CODE RED at this time.
Give me some time to look over the info and share what your battery is doing and hopefully why.
 

I think you are over analyzing things ,those figures look fine. 3,6V is not anything crazy ,there are batteries that only reach 100% soc when a cell hits 3,6V...  Deye batteries do this as example.   If the battery pulls full capacity when you discharge 95% without any issues i don't see the issue . As long as you are not keeping cells at 3,6V for weeks on end ,that's what causes damage.

I want to help you have better control over your battery so that it does what you want it to do.
This starts by knowing what your battery settings are.
For example, my battery balance start voltage is 3.36V per cell which is too early for me. With a target of 3.5V per cell charge it started to discharge both the high cells and the low cells which made my 75mV delta V into 200mV in a matter of days. So I had to reduce my charge voltage closer to the balance voltage to correct this (which also took days). When delta V was 20mV I returned to the previous 3.5V charge per cell.
So lets get to know your battery...
config  =  battery settings
bat  =  battery status including which cell is in balance mode
This info will help to customize a fix for the battery
BTW nothing meaningful (re balancing) can happen in the flat part of the curve so it is ok to use your batteries at night.

See you have now posted new photos.
I agree that the cells are badly matched and i do think pylotech should seriously look into it.

I would say it all looks pretty normal , it can take many cycles for that lower cells to come up to the level of the others. When you have batteries of different ages the stack will almost always behave like this in my experience. Don't charge to lower than 53,2 as this will hinder the balancing speed and process. 

I was obviously not clear: The overshoot is a concern due to the tendency of the Voltronics inverters to overshoot the target.  When your target is as high as mine (53.0V) the inverters (I have 3 in parallel) will easily overshoot to 54V (the battery cutout) if charge current is set at say 120A.

Yes, that happened from day 1. I’ve added one new module to the bank every 3-4 months later, and always the new module became the master. But that’s the information us5000 communicates as well, that it should be charged to 53.3V. 
During a normal day it would reach 100% at lower voltage, even 51.5V from what I remember, stop charging but later in the day it would charge a bit more and eventually reach 53.3V. That is 3.55V per cell, not too dramatic. However I prefer lower. That’s why I went with voltage and 52V.5V bulk and 52V float.
Sort of unrelated to the 53.3V thing, but even with the 52.5V setting, I do have a sort of an issue I believe with the newest module, after 92-95% one of the cells would imbalance to a difference of 0.120V: highest ca reach 3.6V, lowest would be 3.48V. The bms temperature will rise, meaning that the balancer is busy. The cell temperatures remain around 20C, while the BMS can reach 40C. The module will stay there for a while, at 98%, the other two modules will be already at 100%. After that while(can be one hour) the module will finally be at 100%, cells would be balanced and the voltage of each cell would be around 3.48V.
No alarms, no records in the logging. Pylontech says it’s normal. I do not consider it to be very normal.
 
 
 
 

Sadly, as so often with Voltronics, it rather depends on which manual you look at.  The Manual I linked to above agrees with what you said.  There is another manual (4KVA/5KVA Parallel Installation Guide) that clearly shows breakers for each inverter on both input and output side.  Take your pick...
I have mine installed as per the King manual - just one AC-in and one AC-out breaker.  No problems so far (4 years)
Having separate breakers on the output can cause an issue: if one trips (or is switched off) the system will detect a current sharing fault and shut down all A/C output.
I quite like having only a single breaker: it removes the (very strong) temptation to "service one inverter while the other can be in service".  (I am trying to give up blowing things up 😅)

That's good. I got a different impression, and didn't want others to get a similar different impression, and end up in trouble.
This is unknown territory for me. I vaguely recall turning off the switch of one of my paralleled inverters and getting an instant fault code. It was some time ago, so I can't recall the details, but if accurate, that would be a problem (assuming it held true for your much later models).
As for @Scorp007's point about consuming 40 W from AC, that would also be a problem. My 2015 models have no AC power supply, so that can't happen in my case, unless the inverter is in bypass mode, and then you'd expect AC-in to power the static load for the electronics, display etc. It sounds like you'd also need to interrupt AC-in when you turn off the switch.
However, you are at the mercy of Voltronic's firmware to ensure that it always does the right thing when the power switch comes back on. Also, when a load suddenly comes on, you likely want the sleeping inverter to power up as soon as possible, but it will be waking up, yawning, performing soft starts of the bus and the DC-AC converter, scratching its bum, and finally contributing to the load many seconds later. I guess by then it will have had plenty of time to synchronise with either or both of AC-in and the signals on the CAN bus / paralleling boards. I get the impression that synchronisation happens via opto-couplers from the master to all slaves via a dedicated wire, rather than CAN bus messages. The CAN bus messages seem to be for sending parameters, fault code information, that sort of thing. I haven't discovered if it checks for inconsistency between AC-in and this synchronisation signal; I would hope that it would.
So I would take this into consideration, along with possibly low gains, before implementing this.
BTW, our fully patched firmware for Dynamic Load Control (turning units to low power) operates more or less by making (the) target machine(s) behave as if their power switch is off. However, the big difference is that with our patched firmware method, which operates on an RS-233 command with unusual values, we can't let the inverter completely turn off, otherwise you can't dynamically turn it back on. On low PV voltage models, the solar charger has its own computer that wakes up when there is some 5 V more PV than battery voltage, and it has a "private line" to the main power supply that wakes up the main computer at sunrise even if the switch is off. That's fine for many circumstances, but not for "wake on significant load".

Just a point on switching one off. It will either draw about 40W from the grid if AC in is on or from battery. 
If you need max 4kW it's always better that each one provides 2kW than 4kW from one and zero from the other. 

The DC-AC converter is only about 15 W (fixed, more losses with higher I²R, i.e. higher loads) on a 5 kVA model. Perhaps 12% more on a 5.6 kW model. I have the ability, due to my fully patched firmware, to turn off one of mine, but I find that it makes so little difference that I have lost the appetite to use this feature. The feature, however, is highly asked-for, by users that don't have it.
Without this patch, however, doing this is fraught with danger, as others have reported. There is and never will be fully patched firmware for the high PV voltage models, so my suggestion is to put it out of your mind.
When you have paralleled units, you must never ever dynamically change setting 28. That's just a recipe for disaster. Changing setting 28 to SNG means that there is no need to synchronise the output with anything else, although it will always synchronise, though importantly not instantaneously, with the AC-in if/when present. I don't see what dynamically changing setting 28 would achieve; probably I am misunderstanding your comment.
When inverters are paralleled, you should always have separate breakers on each output. In Australia, we only break the active; I don't know how it would work in a country where it is expected (mandated?) to break neutral as well as active. I would never parallel inverters before checking with a multimeter that the voltage difference between AC-out active terminals is low. To do that, the AC-out neutrals have to be connected.
My Parallel Installation Guide does not show AC-out breakers at all in their diagram, I assumed that this was to reduce the clutter. It does show just one AC-in breaker, and indeed my system has only the one, but I sometimes wish I had separate ones.
I disagree with Voltronic on this one. A breaker is cheap compared to the peace of mind it brings, at least to me.

It was the first 5kVA -4kW Axpert inverters. Always switched the slave off. There was no down time for AC - out.

I do exactly this with my 3 Kings - 2 of them have relays in parallel with the switch that my supervisory system turns off when loads and PV are low.  This has worked fine now for 4 years.
If memory serves me well @Coulomb has released firmware (for my Kings at least) that allows the inverter output phase to be switched off using a serial command.  I have not (yet) tried using that.

2019 I had 2  5kW inveters in parallel. All was fine but the night time consumption was to high. We were running leadacid batteries at the time.I decided  to switch of 1 inverter at night to lower the consumption. After a couple of days when I switched  the inverter it blew 8 of the 16 MOSFETs. I dont know why. I may be wrong but I will not do it again.
 

Have a look at
 

That sounds like the premature float bug to me. Is this on solar charging, or utility charging, or both?
It should reach the bulk voltage, and stay there until the charge current drops to a low level; that's the absorb stage. A bit of a misnomer for lithium batteries, but it's still very important. Without that, the battery won't be reaching 100% SoC.
When you set the battery type to USE, then the battery's BMS is no longer in control, and it's all strictly voltage based.

Most likely it's just capacitive coupling between the cables and the conduit. Earth it and forget about it.

I have done the same on another inverter (Easun iGrid SV4), but I've used two fans, one on each side. They are blowing air inwards as the original fans are not reversed, so the air is still circulating in the opposite way of natural convection :(. The inverter still runs pretty warm, morning temperature is around 33C yet I've never seen it go above 60C since. The ambient temperature is 22-27C. I would not necessarily call this mod a success (I have expected more) however it more of peace of mind in case the oem fans stop for some reason, there is still something on the forced cooling side operational. I've used two Arctic Cooling P8.

Everything BritishRacingGreen said above is true. In my observation, in SBU mode the grid gets connected when the battery will reach the back to grid voltage. However this does not need the battery to be flat. In my case, 48V battery, Pylontech us5000x2. 100% can be reached as low as 51.5V. I have set the back to grid at 48V which can be touched if all of a sudden there is 70%-80% load on the inverter, like the water pump starting when the dishwasher is running, even if there is sufficient PV. The grid is connected briefly, the inverter switches from Solar/Battery to Hybrid (as indicated by Solar assistant), then when the surge has passed it will linger for a while (1-2mins) in Hybrid despite the voltage rise to above 50V (back to discharge setting in my case is 49V), and finally back to Solar/Battery.

BYPASS term has become a bit of a misnomer nowadays. Because the later machines are actually on-grid topology (bidirectional inverter), the grid is  connected to the load even in SUB mode while PV may be supplying the bulk or even the entire load. This is the topology required for AC blending where the gid is online in order to add the shortfall when pv cannot cope with total load requirement. Example a 4kW load gets 2.5kW from pv, and simultaneously 1.5kW from grid. So thats not entirely BYPASS mode, but the legacy terms has been retained. 
So grid is always connected in USB (hard bypass) and SUB ('partial' bypass)  and disconnected in SBU mode. In latter mode the grid will connect automatically though if both pv and battery are not available anymore, typically when battery becomes flat. 
Note also because the machine is bidirectional, the grid CANNOT charge your battery when in SUB and SBU modes, even if charge priority includes grid. 
 

I’ve installed the capacitor two days ago. The problem of the LCD is solved. I’ve waited before reporting back just to be sure. We’ve had days with sun and clouds and those high irradiation when the sun comes out of the clouds, so big stress on the unit. The lcd indications did not even budge. I’ve used a film cap 2.2uF, X2, 630V DC by Kemet. As I only have 240VOC per string this should pe plently. A 3.3uF would also fit behind the screen. I’ve installed it as indicated in the document. Thank you very much Coloumb!