Transistors reach their maximum temperature in less then a minute, so they do not heat up any further
In many cheap UPSes the heatsink on the transistors is just a massive cube of aluminum, and does not have any fins. This makes me think that it is simply meant to absorb the heat with its thermal capacity, rather than dissipate it, and that it is meant to just slow the heating untill the battery depletes - therefore problem with larger battery
Then indeed, their temperature would rise slowly. But on many of them the transistors dissipate so little, then they use only ~1mm thick aluminum plate or nothing at all as a heatsink, what is quirte fast. The reason may be, then the cube is not for free and occupy quite large space, so lower Ron types allow to reduce the UPS size
But the transformer take about a 1 hour to settle it's temperature, so with the high output power the final temperature would be way above the material could withstand
Then you can just load the transformetr less, to lower its heat output. I dont know for sure how much less, but i guess that (output VA / transformer core cross section area) is the thing to look for
If so, can i measure (VA / area) in a quality UPS meant for long work, and use this value for the wanted max. (VA / area) in the cheap UPS ? (adding a fan to blow on the transformer, like the one that exists in the good UPS)
Generally yes, then it apply for transistor dissipations too. But there is quite significant part related to transformer core magnetizing and output capacitor charge/discharge currents, what is there even when unloaded (or it get worse, when unloaded, so the effect of unloading is not as strong)
Usually the overheated transformer start to saturate it's core (it is usually designed with virtually no margin), what heavilly increase the load on transistors, leading to their destruction.
How temperature is related to core saturation ?
On most magnetic materials the saturation flux decrease on higher temperature, so if the system is designed just on the border when cold, it cross it when it heat up significantly.
But for magnetic ballasts you would need really sinewave output UPS, as the "modified sine" (alternating positive and negative pulses with peak value equal to the nominal mains sinewave) have the gaps with zero voltage too long, so following arc reignition (with the next active pulse) is problematic
I tried some combinations, and here are the results (all tests with 240V 50HZ nominal) :
Preheat 36W T8 on square (or modified with just 1 "step") wave = works ok but starter keeps glowing quite bright and is very warm after some time
Preheat 36W T8 on modified wave with 3 "steps" = works ok
Preheat PL 11W on square (or modified with just 1 "step") wave = works ok, i think the starter is glowing and heating inside
HPS 70W Internal ignitor on square (or modified with just 1 "step") wave = cycling
HPS 70W Internal ignitor on modified wave with 3 "steps" = works ok
HPS 150W external self-timing ignotor on modified wave with 3 "steps" = works ok
Bug zapper on square (or modified with just 1 "step") wave = electric net continuously arcing, lamps work ok
Bug zapper on modified wave with 3 "steps" = works ok
I have no experience with "modified sine" with more then 1 step, but indeed, more steps should help sognificantly.
But I doubt multistep modified sine UPS are still made (maybe except some special uses), as by adding steps they are becoming quite complex (multiple transformer windings or converter output levels, lot of power transistors - 3-level one require at least 10..12 power FET's with quite complex predrive).
Today is way simpler and cheaper to make simple (unregulated) true sinewave. With the use of simple microcontroller with PWM modulator it is the same simplicity as mains frequency transformer-less 1-step modified sine - single DCDC generating ~350V, one output H-bridge and a filter - so total 6..8 power FET's with simple predrivers.
