Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

concerned inverter change won't solve production problem

This site may earn commission on affiliate links.
I'm hoping some the folks here can weigh in on whether they think Tesla's recent suggestion to swap my Delta inverter to improve my solar roof's production makes sense.

Apologies for the long-winded description ...

I have a 9.94 kW Tesla solar roof with a Delta M8-TL-US 8kW inverter that was installed in December 2020.
The roof's 170 PV tiles are connected in 6 strings that are joined into 3 parallel string pairs.
More specifically, the strings are organized as follows:

The PV1 parallel string pair has:
- 36 tile string with tiles on south facing (azimuth 167, pitch 40) roof
- 36 tile string with tiles on south facing (azimuth 167, pitch 40) roof

The PV2 parallel string pair has:
- 28 tile string with tiles on south facing (azimuth 167, pitch 40) roof
- 28 tile string with tiles on east facing (azimuth 78, pitch 45) roof

The PV3 parallel string pair has:
- 21 tile string with tiles on west facing (azimuth 258, pitch 45) roof
- 21 tile string with tiles on west facing (azimuth 258, pitch 45) roof

My roof's production has been lower than expected since install. I was originally concerned that misplaced PV tiles might be contributing to the poor production (see Incorrectly Placed Solar Roof Tiles) and then later was concerned about the PV2 parallel string pair which combines tiles from different facing roof planes. In August 2021, Tesla sent a crew out to debug/resolve the production issues and they discovered that my PV1 and PV2 strings had been crossed. So the PV1 pair had a 36 tile string paired with a 28 tile string and the PV2 pair also had a 36 tile string paired with a 28 tile string. The crew fixed the pairing and it now matches the original design listed above.

My roof's production improved, but one year later production was still only 74% of Tesla's original estimate of 11,112 kWh per year. In late July 2022, I asked Tesla to investigate why production remains low. A week or so ago, I learned that they plan to swap my Delta inverter. I don't know whether the inverter is to be replaced with a similar model, a different size Delta, or a different brand (e.g. Tesla), but after looking at
data from my current inverter, I don't see how swapping it out will help since there is no clipping.

Here is an aggregated screen shot of my inverter's data from one of my better (recent) production days:

2022-09-23-stats.png


Does anyone see anything in this data that indicates the inverter has a problem that would limit production?

I'm thinking that the PV2 pair with different oriented strings is a more likely culprit, but am not sure.
Maybe Tesla just made a poor initial estimate.

Any ideas or thoughts are welcome - thanks!
 
Something doesn't add up on the first row of graphs you posted. The string voltage V_mp does not vary that much with irradiance (while the string current, I_mp, varies roughly linearly with irradiance). The first row of graphs appears to show:

PV1: 36 tiles, ~280V peak
PV2: 28 tiles, ~ 290V peak
PV3: 21 tiles, ~ 210V peak.

I'm not clear on the specs on any individual tile, but V_mp should be proportional to the number of tiles for equal irradiance. [Peak irradiance over a day for strings at different orientations will differ, but since V_mp doesn't vary much with irradiance, peak string voltage per tile should be almost the same for different orientations.]

So if you assume that PV2 and PV3 are properly installed and have the proper tile counts (they are both around 10V peak per tile), then PV1 appears to be behaving more like a (parallel pair of) 27 tile string(s), rather than 36 tile string(s).

Cheers, Wayne
 
Last edited:
Thanks Wayne!

I'm not clear on the specs on any individual tile, but V_mp should be proportional to the number of tiles for equal irradiance.
...
So if you assume that PV2 and PV3 are properly installed and have the proper tile counts (they are both around 10V peak per tile), then PV1 appears to be behaving more like a (parallel pair of) 27 tile string(s), rather than 36 tile string(s).

You are correct: V_mp for my tiles is 10.99.

It is interesting that you suggest PV1 is behaving like a 27 tile string because that sounds like the problem Tesla "fixed" in August 2021. I think you are saying that paired string peak voltages should be proportional to the number of tiles (36-28-21 or 100%-77%-58%) - right?

I don't have September 2020 inverter data (from before Tesla's August 2021 fix), but this March 2021 data might still provide a useful comparison:

Screen Shot 2022-09-25 at 1.51.42 PM.png


In this data, PV1 and PV2 current differ significantly.

I'm curious if this looks to you to be consistent with what Tesla said was fixed.
I'm also interested if you think changing the inverter would correct any of this - I think you're suggesting the problem is wiring and/or tiles up stream of the inverter.

Thanks again for your help!
 
It is interesting that you suggest PV1 is behaving like a 27 tile string because that sounds like the problem Tesla "fixed" in August 2021. I think you are saying that paired string peak voltages should be proportional to the number of tiles (36-28-21 or 100%-77%-58%) - right?
Yes.

I don't have September 2020 inverter data (from before Tesla's August 2021 fix), but this March 2021 data might still provide a useful comparison:
Those voltages look like they are in the correct proportions, unlike the first set of graphs you posted.

As to changing the inverter, I don't see how that's going to help. It's not like it's clipping or anything. Perhaps there is some subtle failure mode I'm unaware of, or perhaps there is some power limit per input that is causing the inverter to shift away from V_mp. That's beyond my knowledge of how inverters work under the hood.

Cheers, Wayne
 
I agree that the PV1 voltage seems to be incorrect, implying that the system still isn't wired correctly. However, the current provided by PV1 (sept 2022) increases sharply at just before 11AM. It roughly doubles. Is one set of tiles shaded until then? Similarly, PV3 shows half current from around 2 to 3PM in both the Sept 2022 and March 2021 graphs. Are these half power events caused by shading? If not, the events are potentially reasons to suspect an inverter malfunction.

I have a M8-TL-US 8kW inverter (with panels). It is at FW version 3.1.17. Previously it was 3.1.7 (I was told in spring 2021 that 3.1.8 was a version expressly for the solar roof) and we did see a tendency for it to take ~15 minutes for the power to return to full value after a passing cloud. 3.1.17 seems to have fixed that behavior, although we find no measurable increase in energy production. If it is true that a solar roof actually requires a different firmware, perhaps Delta is no longer willing to support Tesla, and that could be another reason to switch inverters.

My possibly biased impression is that Delta inverters are more reliable than either SolarEdge or the fairly new Tesla inverters, so I would probably be resistant to changing out the inverter until the string voltage question is understood.
 
You've focused on voltages - do the the differences in current/amps reveal anything useful?
Yes, good question. To first (limited) approximation, the string voltage will tell you the number of panels in the string, and the string amperage will tell you the irradiance. And then if you parallel two equal length strings, the amperages will add.

Looking at the March 2021 numbers, you can see that PV1 and PV3 amperages are near half of that of PV2 or the September 2022 numbers. The two different time periods should be comparable as they are both near an equinox.

Thus my current (so to speak) diagnosis: In March 2021 PV1 and PV3 had one each of their parallel strings non producing (or very low producing, perhaps due to a string length mismatch), while PV2 was maybe OK. And in September 2022, the currents are better, you definitely have both string producing on each input most of the day, but there is still some mismatch, as the voltages show no sustained evidence of a 36 tile string.

Correction to that last statement: Briefly in the morning, from 7 a.m. to 9 a.m., PV2 is now behaving like a single 36 panel string, both in the voltages and currents. So I'm thinking you still have a 36 panel string and a 28 panel string mixed up on each of PV1 and PV2. PV1 also show a big jump in current at 11 a.m. which could be explained by the input transitioning from one string producing to two strings producing.

Cheers, Wayne
 
  • Like
Reactions: trautmane2
However, the current provided by PV1 (sept 2022) increases sharply at just before 11AM. It roughly doubles. Is one set of tiles shaded until then?

Hmmm ... maybe I have PV1 and PV2 backwards where PV2 is the 36+36 all south and PV1 is the the 28 south + 28 east. The east facing tiles on my garage could be shaded by my neighbors' (very close) house in the earlier morning hours.

Similarly, PV3 shows half current from around 2 to 3PM in both the Sept 2022 and March 2021 graphs. Are these half power events caused by shading?
I think the PV3 drops are caused by either my other neighbors' house or my house shading the west facing tiles on my garage.

I have a M8-TL-US 8kW inverter (with panels). It is at FW version 3.1.17. Previously it was 3.1.7 (I was told in spring 2021 that 3.1.8 was a version expressly for the solar roof) and we did see a tendency for it to take ~15 minutes for the power to return to full value after a passing cloud. 3.1.17 seems to have fixed that behavior, although we find no measurable increase in energy production.
Yeah - I remember the discussions about that problem. My current Delta firmware version is 3.1.16 and I'm pretty sure I never experienced the cloud problem.

My possibly biased impression is that Delta inverters are more reliable than either SolarEdge or the fairly new Tesla inverters, so I would probably be resistant to changing out the inverter until the string voltage question is understood.
That is also my concern. Without being able to contact the tier 2 folks making the technical decisions and get more specifics, it is really hard to know whether the decisions are sound. I also don't know how much of the low level Delta inverter information Tesla has access to remotely. I wonder if they can even see these stats or if (like for me) the only way to get the low level info is to be standing right next to the inverter.

Thanks for your insight!
 
To follow up on my last post, if your red textual labels are correct, and there's no shading, I would expect voltage curves like the blue lines I marked up below, and amperage curves like the green ones. [The green ones I drew free hand, so minor deviation from the black on PV1 and PV2 around 3:00 p.m. is just my drawing skill, not intentional.]

I'm guessing that PV1 and PV2 both still have a 28 tile string and a 36 tile string in parallel. The deviation from ideal on the PV3 charts could just be shading.

Cheers, Wayne

2022-09-23-stats.png
 
  • Informative
Reactions: trautmane2
Thanks Wayne - very helpful to see.

if your red textual labels are correct, and there's no shading,
I may have PV1 and PV2 backwards - still not sure - but there is shading on the 28 tile east string in the morning. This is a picture of the shading on June 23, 2021 (so the times will differ in September) but it shows how the peak of my neighbors' roof affects things:

2021-06-23-mp2-east-shading.png


The PV3 west facing strings definitely get shaded in the afternoon from my other neighbors house - I don't seem to have a photo of that.
 
I may have PV1 and PV2 backwards - still not sure - but there is shading on the 28 tile east string in the morning.
OK, let's stick with your names PV1 and PV2, and consider that the textual description may be off. Judging by the afternoon current graphs, PV2 has the east facing string, I think.

On September 23, PV1 is showing a shift in behavior just before 11 a.m., shifting from looking like a single string of 28 to two parallel strings of 28; and then it shifts back around 4 p.m. I'm thinking this means it has a 36 tile south in parallel with a 28 tile south.

PV2 is showing a shift in behavior around 9:15 a.m., shifting from looking like a single string of 36 to two parallel strings of 28. I'm thinking this is the 28 tile east plus a 36 tile south. The behavior is due to the east tile shading; when it's heavily shaded, it's optimal to ignore that string and just run the 36 tile string; but when it's less heavily shaded, it's optimal to run both strings at the highest voltage the 28 tile string can handle.

Cheers, Wayne
 
PV2 is showing a shift in behavior around 9:15 a.m., shifting from looking like a single string of 36 to two parallel strings of 28. I'm thinking this is the 28 tile east plus a 36 tile south. The behavior is due to the east tile shading; when it's heavily shaded, it's optimal to ignore that string and just run the 36 tile string; but when it's less heavily shaded, it's optimal to run both strings at the highest voltage the 28 tile string can handle.
Yeah - that makes sense to me. Thanks for puzzling through it all!

I'm left wondering what the original (pre-August 2021) wiring was, since the current arrangement, while flawed, is still 20% better than the original.

From the beginning, I was always bothered that the tiles were not placed exactly where the design specified. I think I'm convinced at this point that not-quite-to-spec tile positions don't matter and that the real issue is just with wiring. I did however map out the south roof PV tiles: where they were originally installed, where they were supposed to be, where 3 tiles got moved during the August 2021 repair, and where I thought the strings were:

2021-08-04-pv-plan-v-actual-south-fix.png


Thought I'd include this just in case it was of interest ...
 
The Tesla maintenance crew arrived a couple days ago to try to fix my production issues. I now know a little more about their plan and the rationale behind it.

The plan is to replace my 8kW Delta inverter and Mid-Circuit Interrupters (MCIs) with a 7.6kW Tesla inverter and Tesla MCIs. Apparently, the Tesla MCIs require less startup voltage and can operate with only some tiles (modules) in each string being active while the Delta MCIs require more startup voltage and require all tiles in the string to be active. So ... the theory is that with Tesla MCIs, my roof will start production a little sooner each morning, stop production a little later each evening, and handle shading conditions better.

The Tesla inverter, other than working with the improved Tesla MCIs, doesn't seem to provide any direct production benefit. In fact, it will clip a little sooner (7.6 vs. 8) but clipping is not an issue for my current setup. The Tesla inverter will be easier for Tesla to maintain, monitor, and replace going forward - so that is also a plus.

As mentioned earlier, I'm a little concerned that Tesla's inverter is pretty new and they might still be working out kinks in its production and design.

I also just saw this post from @tpapadopoulos :
Tesla replaced my MCI units because of poor production in the winter, but now it's much worse.

which worries me because it looks like a similar situation to mine that did not work.

Fingers crossed ...
 
I really hope it helps. I have had 2 of the 7.6 Tesla inverters and they have been working great for the last year. I can’t speak to whether this change will help you or not because I am not an expert. Just wanted to chime in one little datapoint about my experience with the Tesla inverters.
🤞
 
My repair project is almost done - hopefully, just one more day of work from the crew early next week.

Screen Shot 2022-10-08 at 10.04.36 AM.png


Even though there is one missing string of tiles at the moment, the crew kindly enabled the other strings and the new Tesla inverter so I would get some production over the weekend. This allowed me to connect to the inverter and unfortunately, realize that the Tesla inverter actually reduces my monitoring functionality. It looks like Tesla views inverter connections as a setup-only thing, so you can only connect within 15 minutes after the inverter has powered-up. Once connected, you only see point-in-time string stats - no historical information. This is all disappointing since after @wwhitney 's posts here, I finally understand what to look for to confirm the strings are behaving properly.

I wonder if anyone with a Tesla inverter has found a way to get historical string data. I have code that uses the Tesla Energy API to pull aggregate data (the same stuff you see in the Tesla app) but I haven't found any endpoints for lower-level string data. Maybe Tesla doesn't even track that? I was always curious whether Tesla could see the string data from my previous Delta inverter remotely.

Sigh ...