Slave and master with 2 Axpert 5Kv

[incagarcilaso]

Hi. I have been running two 5Kv Axperts in an installation for some time now with no issues, but just recently there are two things that I have noticed and am not sure if they are faults or normal behaviour:

1. Scenario: they are running with more input from solar than load (cloudless day) and then a large load kicks in which pushes the load over input by quite a lot and so the batteries have to kick in. When this happens I see that the inverters suddenly show 0 input for a few seconds which then slowly climbs back up to its previous amount. Is this correct behaviour? Why do the inverters block input when this swing happens? It only lasts for about 30 seconds but it does mean that during that time the batteries are under a lot more pressure than they should be (for example if the input is 3000 W and the load is 5500 W the batteries have to cope with a load of 5500 W for 30 seconds rather than just a 2500 W load). In some cases this might result in the inverters switching to bypass unnecessarily (because they have to carry 5500 W instead of just 2500 ").

2. The inverters sometimes switch roles between slave a master. I hadn't noticed this before and don't think it has happened before now but their roles have suddenly reversed. The master has become the slave and the slave the master, without any other apparent issue.

Thanks for any input/ideas.

[___]

23 minutes ago, incagarcilaso said:

When this happens I see that the inverters suddenly show 0 input for a few seconds which then slowly climbs back up to its previous amount.

I don't know the Axpert that well but I see something similar with my Victron equipment. The MPPT will occasionally go hunting for alternate power peaks. Without getting too technical, it is sometimes possible that in a combined array there might be multiple local maximae and with the usual perturb and observe method used by the majority of MPPTs it is impossible to find those without doing a "rescan". To do a rescan the MPPT effectively swings the voltage across the scale and checks for power delivery. The BlueSolar units does this three times... so you'd stand there wondering why on earth it goes down to half power with no clouds in the sky... well that is why.

Now, I noticed that on sudden load changes, usually when the load ends up balancing solar production, that the MPPT will pick this voltage change event as the perfect opportunity to go do a scan. So the effect is similar: Large load starts, and suddenly the MPPT goes scanning leaving the batteries to do the work.

It is possible that something similar is happening with the internal MPPTs in the Axpert. The Victron's are blazingly fast, this process takes all of 5 seconds, so I cannot conclusively say if 30 seconds is "normal", but given a naïve perturb and observe algorithm it is certainly possible that it might take that long to readjust.

I wonder if the other Axpert owners have seen similar behaviour?

[ibiza]

exactly the same here!

[incagarcilaso]

4 minutes ago, plonkster said:

It is possible that something similar is happening with the internal MPPTs in the Axpert. The Victron's are blazingly fast, this process takes all of 5 seconds, so I cannot conclusively say if 30 seconds is "normal", but given a naïve perturb and observe algorithm it is certainly possible that it might take that long to readjust.

I wonder if the other Axpert owners have seen similar behaviour?

This is a great help and I imagine that your guess is correct. It puts my mind at rest although it will not solve the problem. This must happen when the load momentarily equals the production as the load swings upwards. I have noticed that this is the point at which it happens, although it does not always do this, only sometimes. The problem with this is that if the batteries are already a little low this sudden drain, even if it is only for a few seconds can take them below the threshold I have set for switching to mains, and I can't realistically put this any lower. As you say it is quite probable that the Axpert's MPPT controller works in a similar way to the Victron.

[___]

It even makes sense I suppose, that it chooses that time to do a new scan. Imagine it's got the batteries at absorb or even in float, so there is enough power coming in from the PV side to get the job done. There is no point in looking for a new maximum point. Suddenly the voltage goes down as a big load starts, which immediately tells the MPPT to increase power. That leaves it with two options: adjust the voltage from where you are, or scan the range again?

The BlueSolar unit seems to do this in phases, so if you watch it you see the power drop to about half, then back up to the former peak, then it drops again to about a third, and then goes up again, then it drops again to an even lower point, and then it finally settles back at the former peak. It does this very quickly.

I suspect this is what the marketing material refers to when it says "Advanced Maximum Power Point Detection in case of partial shading conditions." :-)

[incagarcilaso]

26 minutes ago, plonkster said:

It even makes sense I suppose, that it chooses that time to do a new scan. Imagine it's got the batteries at absorb or even in float, so there is enough power coming in from the PV side to get the job done. There is no point in looking for a new maximum point. Suddenly the voltage goes down as a big load starts, which immediately tells the MPPT to increase power. That leaves it with two options: adjust the voltage from where you are, or scan the range again?

Yes, has to be this. It seems, then, that maybe the Axperts don't handle this procedure quite as well as the Victron (from what you say) because it takes them longer to run the process but the difference I suppose is negligible. The time I am seeing for it to run the procedure is coming in from a third-party app as well so it may be that it is not taking this long to actually do it.

Any ideas on issue 2, why the switch between slave and master happens with 2 Axperts in parallel? I wonder if it is related to this issue of scanning and it is at that point that the switch takes place?

[Johann1982]

2 hours ago, incagarcilaso said:

Yes, has to be this. It seems, then, that maybe the Axperts don't handle this procedure quite as well as the Victron (from what you say) because it takes them longer to run the process but the difference I suppose is negligible. The time I am seeing for it to run the procedure is coming in from a third-party app as well so it may be that it is not taking this long to actually do it.

Any ideas on issue 2, why the switch between slave and master happens with 2 Axperts in parallel? I wonder if it is related to this issue of scanning and it is at that point that the switch takes place?

@incagarcilasoI had 2x Axperts in parallel before. The only reason it would change the master and slave around is when you shut them both off and restart. It automatically chooses the master inverter and it can differ from time to time with complete shutdown and restart. Usually when batt voltage drops out completely and there is no mains.

[incagarcilaso]

2 hours ago, Johann1982 said:

@incagarcilasoI had 2x Axperts in parallel before. The only reason it would change the master and slave around is when you shut them both off and restart. It automatically chooses the master inverter and it can differ from time to time with complete shutdown and restart. Usually when batt voltage drops out completely and there is no mains.

Yes, I had understood this was the case, which is what worried me as to may knowledge they have not been restarted. There may have been a power cut but the batteries have never been disconnected so there would always have been power to the inverters. Odd.

[Coulomb]

I suspect that their control algorithm has integral wind-up. This causes delays and overshoots. Of course, it could also be poor PI coefficients, or both (poor coefficients and integral wind-up).

Also, when it shows PV charge current as zero, it means there isn't a net charge to the battery.  It's still supporting the load, which you can verify on the LCD by pressing the down button about 5 times to see the battery discharge current. Also, the PV power (as opposed to PV charge current) will show non-zero power, in fact the same as before the sudden load. So the problem is that it's slow to increase the PV input, not that the PV input goes to zero. You should be able to verify this with monitoring software.

I have a Blue Sky Energy charge controller beside my Axpert, and it's very noticeable how much quicker to act it is (BSE compared to Axpert), and how much lower the over- and under-shoots are.

[SilverNodashi]

16 hours ago, incagarcilaso said:

Hi. I have been running two 5Kv Axperts in an installation for some time now with no issues, but just recently there are two things that I have noticed and am not sure if they are faults or normal behaviour:

1. Scenario: they are running with more input from solar than load (cloudless day) and then a large load kicks in which pushes the load over input by quite a lot and so the batteries have to kick in. When this happens I see that the inverters suddenly show 0 input for a few seconds which then slowly climbs back up to its previous amount. Is this correct behaviour? Why do the inverters block input when this swing happens? It only lasts for about 30 seconds but it does mean that during that time the batteries are under a lot more pressure than they should be (for example if the input is 3000 W and the load is 5500 W the batteries have to cope with a load of 5500 W for 30 seconds rather than just a 2500 W load). In some cases this might result in the inverters switching to bypass unnecessarily (because they have to carry 5500 W instead of just 2500 ").

2. The inverters sometimes switch roles between slave a master. I hadn't noticed this before and don't think it has happened before now but their roles have suddenly reversed. The master has become the slave and the slave the master, without any other apparent issue.

Thanks for any input/ideas.

This is quite normal on the Axpert's, since they're not really hybrid inverters. Generally, grid tie inverters work better in this case, i.e. they will take the excess load (2500w in your example) from the batteries / eskom (depending on availability and settings) 

[Don]

16 hours ago, plonkster said:

I wonder if the other Axpert owners have seen similar behaviour?

This happened to me twice the day before yesterday. Initially some clouds move across the sun and therefore a drop in PV Output. Then the PV watts dropped to 0 like a dead bird out of the sky and 5 seconds later start picking up again and go back to where it should be. I have added 2 graphs. The one showing the PV Watts only and with the second I included the Load and Battery Watts as well.

[Fritz]

I have the same problem/concern. 

The input PV watts drop to 0 and then picks it up again. The batteries have to take the full strain of the output. in my case the geyser. (45A DC @ 48V)

Why cant it just keep the the available PV input to the load. Surely it must sense there is a bigger load connected at that point. 

I dont like it.....   :-(

 

 

[___]

I'm attempting to reproduce this and capture the CCGX output. Will post when I manage to do it :-)

[___]

Did it.

I start out with the batteries at absorption voltage, a condition I provoked by limiting the maximum discharge to 50W. I then crank the discharge up to pretty much what is available at the moment, forcing the MPPT to increase the power. You will note how the PV charger power dips down 4 times before it settles back on 670W (ish).

Also a nice show-case as to what you can do with a Multi :-)

out-1.ogv

Edit: Lower quality youtube version

 

[Guest]

Nice Plonster

Sent from my SM-G920F using Tapatalk

[incagarcilaso]

5 hours ago, Fritz said:

I have the same problem/concern. 

The input PV watts drop to 0 and then picks it up again. The batteries have to take the full strain of the output. in my case the geyser. (45A DC @ 48V)

Why cant it just keep the the available PV input to the load. Surely it must sense there is a bigger load connected at that point. 

I dont like it.....   :-(

 

 

Exactly what happens in my case and your point about keeping available PV was also why I thought it was an error. It can represent a fairly serious problem if it causes it to keep switching to mains on a day of intermittent cloud. What's more over time these sudden large drains on the battery bank can't be good. @Don 's graph reproduce exactly the behaviour of my Axperts - drops like a dead bird out to the sky, and also takes a good half minute to pick up again.

[incagarcilaso]

@plonksterOK, so this demonstrates how much, much better your equipment handles this scenario than the Axperts. In the same scenario with the Axperts, the PV array input doesn't just dip by 150-200 Watts a couple of times - it plumits to zero, holds it there for 7 seconds and then takes another 30 seconds to gradually restore from zero to the full input of the PV array. This is a heavy toll on the battery bank, especailly if you are running a 5Kw load with 4 Kw PV array input. This can happen on a daily basis if, say, the washing machine and oven go on at the same time. It also means that when your input is really good and the whole system should only put a 20 amp strain on the batteries for a while, you have a sudden 100 amp strain (48V system) on the batteries lasting for almost a minute and returning to a 20 amp drain. For many systems the battery bank simply won't cope with this kind of load and even for battery banks that can, it's likely to shorten their overall life. 

This is something I did not know about the Axperts and is an eye opener. Of course these are only features that you can find out about through use and will not appear in the specs.

[___]

Fwiw, the Microcare MPPTs does the same. Every now and then it rescans the voltage range, and when it does so, it also drops the "loads" like a hot potato.

I'm unsure how the MC does it. When I had it on the bench,  I had a current meter in circuit. I was expecting it to "do a trial run", effectively sweep the mark-space ratio of the buck converter over the range while measuring output power, but that is not what it does. When you turn it on, the screen shows that it's tracking the maximum point, then it indicates the voltage it's decided on, and only then does the current start to flow. So it does something else, no idea what. It has some kind of logic as to when it decides to redo this. When it does so, power goes to zero.

Edit: I wrote this post before @incagarcilaso's response above... just to show I'm not trying to be unnecssarily harch on the Voltronics :-)

[Coulomb]

Ok, I think I misunderstood. For the PV power to crash to zero only to recover over the next 30-60 seconds or so is an undershoot of the SCC (Solar Charge Controller). I think this could be explained by integral wind-up.

So I'll assume a standard PI control loop: P for proportional, I for integral. You have a set-point, desired PV current, which the MPPT algorithm, whatever it is, has set. You have a current measurement, and the difference of these is the error. You are trying to drive the error to zero. The P coefficient is multiplied by this error to adjust the present PWM ratio. So if the present current is too low, you up the PWM (pulse width modulation), and if it's too high, you reduce the PWM ratio. There is the complication that if the predicted maxim power point current is wrong, you could go the wrong way, but let's ignore that and assume it's pretty good.

The classic problem with P only control is that there is always an error. You can reduce the error by increasing the P coefficient, but then you risk overshoots, undershoots, and instability, possibly even oscillations.

So you use an integral term, I, which introduces a change proportional to not the error, but to the integral of the error. This is usually as simple as a running sum of rhe error terms. So if the PWM value is a little too low, rather than having a little too little current forever, the integral of the error starts accumulating, and this larger and larger galue multiplied by the I coefficient is added to the PWM value, and in the steady state the error can go right to zero. Good.

The problem is when the integral  of the error grows large and the system is near its limit. So: sunny day, moderate load, high PWM ratio, PV output is near maximum. There has been a steady error that the I term has been correcting, so the sum of errors value is high. Now we get a dark cloud, the MPPT current instead of being near maximum is now say 20% of what it was, so for maximum PV we need a much lower PWM value.  At the present PWM value, we're hitting the panels way too hard, so their voltage collapses beliw battery voltage, and we get zero PV current. The bird has dropped from the sky.

Again, it might take the MPPT algorithm a while to realise this, but let's say it's pretty good snd has nearly the perfect value 2 seconds after the cloud starts. Now we have a large positive error (too little current), but it's because the PWM value is way too high. On top of this, our sum of errors value is high, forcing the PWM value too high.

Integral wind-up happens when the sum of errors value is too high, and the best thing to do is to zero it, or at least truncate it. I think they're not recognising this situation, and this causes a long delay while the error heads in the right direction, but the sum of errors term takes up to a minute to finally cslm down and stop making things worse.

You can tell I've never implemented an MPPT control system, but I've done a few standard control systems snd integral wind-up has often been a problem.

Maybe they don't bother with I terms at all (just P on its own) when you have the MPPT complication. In that case, they might have a slack MPPT algorithm, which takes a long time to react to a change in insolation. For example, they might just do small purturbations either side of the present setting, which tracks steady and slowly changing conditions well, but takes a long time to adapt to abrupt changes in insolation. Maybe they do a full scan (try every PWM ratio and pickvthe best), but only do irlt oncrle a minute because thst spends a lot of time away from the ideal setting.

Tricky stuff. The more expensive gear will have more time invested in getting the algorithms right.

[___]

It is of course possible that when the MC shows "Maximum Power Point...." on the screen that is just for show, and that what it really does is measure the open circuit voltage and then dive in around 85% of that value, doing a P&O from there (perturb and observe).

[incagarcilaso]

3 minutes ago, plonkster said:

Edit: I wrote this post before @incagarcilaso's response above... just to show I'm not trying to be unnecssarily harch on the Voltronics :-)

Very fair. The Axpert is a great inverter, relation price, spec, performance. I'm happy with mine, just a shame about this issue. I'm sure they'll iron it out in new models or maybe it's something they can do with a firmware update, not sure.

[___]

2 minutes ago, incagarcilaso said:

with a firmware update,

Exactly, that's all it would really take :-)

[Don]

4 hours ago, plonkster said:

Exactly, that's all it would really take :-)

Assuming, they are aware of the problem. 

[___]

4 hours ago, Coulomb said:

You can tell I've never implemented an MPPT control system,

Frorm what I've read the control loop is way simpler than that. You pick a starting point, and then you change the duty cycle of the PWM slightly and monitor the change in power. If it comes down, you change direction and modify the pwm in the opposing direction. If the power increases you keep going in that direction, when the power decreases you switch direction. When you hit the MPP you basically end up balanced right there on the cusp. This is known as perturb and observe.

The other method is called incremental conductance. As far as I can see this uses the mathematical fact that at the MPP, dP/dV=0. So by looking at whether dP/dV is positive or negative, and how big it is, you know both in what direction the power point is and you can estimate whether it is close or not. To measure dP/dV (or dI/dV which gives the same result really) still involves slightly perturbing the PWM duty cycle to create a delta(V), take two measurements to get a delta(P), and then delta(P)/delta(V) approximates dP/dV.

So if dP/dV is far away from zero, it means you can increment your PWM cycle with greater increments, which allows the tracker to find the MPP quicker.

If the MPPT wants to make less power, it pushes the voltage up (decreases the PWM cycle) so that the modules deliberately run at a non-optimal voltage. In theory, when an increase in power is required, all it has to do is back the voltage down back to the MPP again.

This is grossly oversimplified though, and once again, from what I've read, there are times when you don't have just one peak but multiple peaks across the voltage range. This usually happens in partially shaded conditions. Unless you have some kind of a plan/algorithm to also try "other" voltages, the only peak you will find is the local one closest to Voc. I'm not qualified enough to say whether this is the reason why these controllers do a "scan" of sorts, but I suspect that is why they do it, and why they occasionally redo it.

As shown in the video, it seems the more expensive units attempt to limit the amount of time it takes to scan, while others essentially start over. I know the Microcare (the one I have at least) starts over every now and then.

On a separate topic, this inability to find peaks outside of the local maximum you're perched on is why I'm somewhat sceptical of "evolutionary" approaches to philosophical questions. Eg Sam Harris' "Moral Landscape". Picture sitting on a hill and being unable to find a higher hill, because the algorithm says that when you find yourself going downhill you should turn around and go back where you came from. This is the same sort of issue you face with MPPTs :-)

[Coulomb]

5 hours ago, plonkster said:

On a separate topic, this inability to find peaks outside of the local maximum you're perched on is why I'm somewhat sceptical of "evolutionary" approaches to philosophical questions. Eg Sam Harris' "Moral Landscape". 

Ok, let's not get too off-topic. I'm quite a fan of Sam's, and listen to most of his podcasts. But I agree with you on this.