The generation curve is clean until about half past one, then it steps down, sits flat for two hours and recovers by four. From March to May it happens on every clear day. In December it never happens at all. Before anybody replaces anything, pull the inverter's event log, because three different faults draw almost the same shape on a graph and they have different fixes.
Three faults, one shape
Temperature derating. The inverter is not faulting, it is reducing output to protect its power stage. The curve rounds off rather than dropping vertically, and the log shows a derating or power limit flag rather than a fault code.
Over temperature shutdown. Output goes to zero, then returns after a cool-down delay, often repeatedly. The log names an internal temperature fault and usually records the temperature reached.
Grid over-voltage trip. Output drops to zero at the same time of day, but the log says AC over-voltage. This is not a heat problem. It is the local network voltage sitting high at midday, plus the drop across your own AC cable, forcing the inverter's terminal voltage past its permitted limit. Replacing the inverter changes nothing.
For overvoltage events, a competent engineer should compare logged voltage, cable impedance and utility conditions. Cable or connection changes may be appropriate; transformer taps and protection settings require the responsible party's approval. Do not widen trip limits to conceal a supply problem.
Where the heat comes from
Request the exact model's operating-temperature range and derating information. Compare the measured air temperature at the inverter with those limits, including required clearances and cooling arrangements.
Direct sun on the enclosure. A unit on a west-facing wall is being solar heated all afternoon. The rating assumed it was not.
Mounted under the array on a metal roof, sitting in a pocket of trapped hot air with no cross flow.
A small sealed room. The room warms to whatever the inverters put into it. Even at high conversion efficiency, a few hundred kilowatts of inverters reject a continuous few kilowatts of heat, and in a closed room with no path out, the air temperature simply climbs until something derates.
Units packed too closely. Manufacturers state minimum clearances above, below and to the sides. Line up more units than the room's air movement supports and the exhaust of one becomes the intake of the next.
Blocked filters and heat sinks. On a jute, textile, rice or cement site the filters clog in weeks rather than months, and fins pack solid with fibre. Bird nests behind a wall-mounted unit are common enough to check for by default.
Low DC input voltage. For the same power the inverter carries more current, and more current means more heat. A string design that lets Vmp sag on a hot roof makes the thermal problem worse at exactly the wrong hour.
What the design should have allowed for
None of this is difficult to get right at design stage, and all of it is expensive to correct afterwards.
Record fault codes, output, ambient temperature and operating conditions before resetting the unit.
Inspect and service filters and fans using the manufacturer's safe shutdown, isolation and cooling procedure; do not assume a filter is safe to change while equipment is energised.
Have cooling designed from equipment heat losses, allowable temperature rise, airflow heat capacity, inlet conditions and actual ventilation resistance.
Check electrical supply and string measurements through a qualified person if the logs indicate a non-thermal fault.
A fault log narrows the investigation; it does not replace measurements or prove the cause by itself. Retain the before-and-after records when a corrective change is made.