Module degradation is the easiest thing to measure and almost never the thing that costs money. How modules, inverters, connections and a forgotten monitoring portal age differently, and which of them produces the unplanned bill.
Twenty-five year performance warranties have trained buyers to treat module degradation as the main long-term risk in a solar plant. It is the easiest thing to measure and it is almost never the thing that costs the money.
Modules lose output in two stages. There is an initial drop over the first days of exposure, then a slow annual decline, and manufacturers warrant the two separately. The warranty curve is a floor rather than a forecast, and modules from a serious manufacturer generally track above it.
Two mechanisms can push a plant below the curve, and both are relevant in Bangladesh. Potential induced degradation happens when high system voltage, heat and humidity combine to drive charge across the encapsulant, and this climate supplies two of those three all year. The defences are choosing modules built to resist it, following the inverter manufacturer's earthing arrangement instead of improvising one on site, and testing for it if output falls away in a pattern that follows position within the string. The second mechanism is microcracking, which comes from rough handling, from people walking on modules and from installing on a rail that is not flat, and it degrades quietly for years.
Even so, if you are looking for the number that hurts, module degradation is not it. It is slow, it is spread across decades, and it was already in the yield model.
Power electronics do not last as long as glass and aluminium. DC link capacitors age with temperature, cooling fans are wear items with a defined service life, and the whole assembly runs hot in a plant room that is hot to begin with. Plan a rooftop plant in Bangladesh on the assumption that string inverters will be replaced at least once inside the life of the array, with fans and filters treated as consumables in the meantime.
That is a manageable cost because it is a scheduled one. What turns it into an unmanageable cost is availability. If the model has been discontinued and there is no service presence in the country, a failed unit becomes a multi-week outage while a replacement is imported, and the replacement may not talk to the existing monitoring. Ask about the manufacturer's service arrangement in Bangladesh and the spares position before choosing the inverter, not after a failure.
Keep the plant room cool, sealed and clean and the inverters last longer. It is the highest-return maintenance action anywhere on the plant, and it costs less than any of the alternatives.
The failures that generate real, unplanned expenditure are almost always terminations.
These cost money three ways at once. Lost generation while a string is down, often for months if nobody is watching string-level data. Damage to the module or equipment at the point of the fault. And risk, because a hot DC joint on a roof is how a fire starts.
The most expensive degradation we find on older plants is not electrical at all. It is a monitoring system that stopped reporting two years ago, because the data SIM was never renewed or the portal login left with an engineer who resigned. The plant carries on generating at whatever level it has settled to and nobody knows the difference. Every fault described above becomes invisible.
If you plan one thing for the long life of a plant, plan for a named person to look at string-level yield every month and compare it with the same month a year earlier. That single habit catches a heating connector, a dead string, a soiling problem and a derating inverter long before any of them becomes a bill.