Inverter aging?

[Clark Thomborson]

Hi, I'm wondering if you have thought about modelling the aging of an inverter, by building a thermal model from first-year data, and comparing this to its thermal behaviour in later years.

I'm asking because I have weak evidence that this will be feasible... but it's rather beyond my paygrade ($0, I'm fully retired 😉) and expertise to do anything more than the simplest thermal modelling.  Somewhat to my surprise, I have Rsq = 0.8 on a linear fit of MPPT.I, MPPT.A, Temperature, and timestamps (mostly at 1 min  intervals) in selected months from 5.5 years on a small household system: Goodwe 3000NS + 6x270W Trina panels.   The inverter is mounted pretty close to the ceiling in an unheated hallway -- which (since I'm in Auckland NZ) rarely gets below 15 degrees C or above 25 degrees C.  As you'll probably know, the Goodwe 3000 NS is convection-cooled, so I don't have to model how a cooling fan affects its dissipation.  Furthermore, with such a narrow band of ambient temperatures, it's not very important to model how they affect its dissipation.   I'm modelling the Goodwe 3000NS as a series resistor of 0.5 ohms and a shunt resistor of 1600 ohms, using the logged MPPT.I and MPPT.A values to estimate its rate of heat generation.

The regression equation (with coefficients from R's lm()) is $\delta  T / \delta t = 2.1 - 0.12T + 0.30w - 0.23\mbox{\rm lag}(w)$, in units of degrees Celsius per hour, with $T$ being the inverter's temperature, $w$ being its modelled heat-generation in Watts, and the lag interval being one hour.  My justification for this model is that (by inspection of the logfile) the modelled junction-heat has a near-immediate effect on the inverter's temperature; with a (very rude!) correction for the dozens of minutes required to reach thermal equilibrium handled surprisingly well by the lagged junction-heat.   With these coefficients, thermal equilibrium is about 40 degrees when the inverter is generating 40W of heat; rising to about 64 degrees when the inverter is generating 80W.

The residuals show *some* indication of tracking upward over time, see below.  I'm thinking that a more accurate thermal model might allow a reasonably-accurate and unbiased estimate of the equivalent series resistance of the inverter; and that it will increase over time as its SCR junctions passivate.  My Goodwe 3000 is so lightly loaded that it doesn't ever overheat (its max temp to date is 61 degrees); but my thermal model suggests it'd get above 70 degrees C (and probably go into a high-temperature shutdown mode) if it were dissipating more than about 90W for a couple of hours, i.e. I'm not confident it could handle much more than 2kVA of PV panel without overheating... despite its nameplate rating of 3kVA.  

All to say that I'd expect inverter-aging (and occasional overheating) to be a fairly important consideration on systems with Goodwe inverters with "normal sized" PV arrays.  My array is so far undersized that PVSyst refused to simulate it... and I guess a Goodwe 2000 would be (arguably) a better choice for my PV panels, even here in NZ with irradiance sometimes above 1100 W/m2, if one weren't particularly worried about overheating and premature aging.

I'd be happy to supply my dataset and R sourcefiles, if you are interested in them.  And: I'm grateful that your firm had allowed me to use PVSyst for a 30-day trial.  What a nice simulator!

Cheers,
Clark

Edited November 6, 2023 by Clark Thomborson
typo

[André Mermoud]

PVsyst doesn't take an ageing of the inverter into account.

On which operating parameter would it be applied ? I doubt that the efficiency could be affected, as this would increase the heating along its life.

The only effect I can see is the reduction of the lifetime (MTBF). This is not addressed in PVsyst in the present time.

Now for the evaluation of the produced heat during operation, the best way would be to use the inefficiency, which is a curve according to the power, and  possibly to the input voltage. This is directly accessible in the simulation data  (IL_Oper loss) for each operating hour.  

 

 

 

[Clark Thomborson]

On 11/7/2023 at 10:49 PM, André Mermoud said:

PVsyst doesn't take an ageing of the inverter into account.

On which operating parameter would it be applied ? I doubt that the efficiency could be affected, as this would increase the heating along its life.

The only effect I can see is the reduction of the lifetime (MTBF). This is not addressed in PVsyst in the present time.

Now for the evaluation of the produced heat during operation, the best way would be to use the inefficiency, which is a curve according to the power, and  possibly to the input voltage. This is directly accessible in the simulation data  (IL_Oper loss) for each operating hour.  

 

 

 

Thanks for your response.  I'd suggest that you consider "aging" your inverter-efficiency model for the same reason that you "age" your photovoltaic-efficiency model.  Every Joule of heat thrown off by the inverter is one less Joule of useful output from that inverter.  However I really don't know whether this is a significant factor for very many of your paying clients.  I'd imagine that any utility-grade inverter, as well as any upmarket domestic-grade inverter, is designed more conservatively (with respect to current density through the junctions of its power transistors) than the Goodwe 3000 NS; and it may be that the  apparently-decreased efficiency of my Goodwe 3000 is an artefact of my modelling.  

You might also want to expand the use-cases for PVSyst, so that it's helpful to the owners of PV systems who are considering replacing their aging inverters, or who are interested in developing a better life-expectancy model.  My websearch suggests this is a growing concern, see e.g.. https://www.utilitydive.com/news/us-solar-farms-are-aging-is-it-time-to-begin-repowering/690978/, https://www.dnv.com/Publications/pv-inverter-useful-life-considerations-144365, and https://ieeexplore.ieee.org/document/9216069