An ordinary grid-tied inverter will not simply work on generator supply. Why it trips within seconds, what reverse power does to an alternator, the minimum loading problem, and the four control options in order of cost.
The first time a factory runs its new rooftop plant on generator supply, one of two things happens. Either the inverters trip within seconds and stay off, or they stay on and the generator starts behaving oddly: hunting, over-frequency alarms, in the worst case a reverse power trip that drops the whole site. Neither is a fault in the equipment. Both are what you get when a grid-following inverter is asked to work on a source that is not a grid.
A standard grid-tied inverter measures the voltage and frequency in front of it and injects current in step with what it finds. It does not set frequency and it cannot form a voltage by itself. It also carries anti-islanding protection, required by the utility and tested to IEC 62116, whose whole purpose is to detect that the source behind it has gone and to disconnect quickly.
Most anti-islanding schemes work by injecting a small deliberate disturbance and watching how the source responds. A stiff utility grid absorbs the disturbance and nothing moves. A single diesel set is a small source with limited inertia, so the disturbance actually moves the frequency, and the inverter reads its own perturbation as evidence that the grid has gone. It disconnects. That is the protection working correctly rather than failing.
Add to that the ordinary behaviour of an engine governor. As PV picks up load, the engine backs off, and a lightly loaded governor holds frequency less tightly. The inverter sees frequency wandering outside its window and trips on that instead.
On the grid, surplus generation flows out through the meter and becomes an export credit. On generator supply there is nowhere for surplus to go. If PV output exceeds the site load, the excess flows back into the alternator, which cannot absorb real power. It motors the engine instead, and the engine is not designed to be driven.
That is what the reverse power relay on the generator breaker, the ANSI 32 function, exists to prevent. If it is fitted and set correctly, it trips and you lose the site. If it is not fitted, you get overspeed, over-frequency and mechanical damage. So the entire control problem reduces to one rule: on generator supply, generation must never exceed load.
The reason factories want solar running through a generator period is fuel. The catch is that a diesel engine run lightly loaded for hours suffers incomplete combustion, unburnt fuel and oil past the rings, glazed bores and carbon in the turbocharger. Generator makers publish a minimum recommended continuous loading for exactly this reason. Read the manual for your own set rather than accepting a contractor's assumption, and treat that figure as a floor.
So the target is not maximum solar. It is as much solar as possible while keeping the engine above its minimum loading and well clear of reverse power. That is a narrower operating band than most owners imagine, and on a large array during a light shift it can mean curtailing a good deal of available generation. Better to know that before the project than after.
On a clear February morning the control problem is slow and easy. Through the pre-monsoon period, when cloud builds through the afternoon, irradiance across a large roof can fall by a large fraction in seconds. The engine has to make that up, and an engine's ability to accept a sudden load step is what limits how much PV you can safely run in parallel with it.
Two settings deal with this: a cap on PV penetration while in generator mode, and a ramp rate limit on the inverters so that recovery after a cloud does not arrive as a step. Both belong in the specification, with figures agreed against the engine's load acceptance data rather than assumed.
An auxiliary contact from the changeover panel inhibits the inverters. It costs almost nothing, saves no fuel, and is entirely safe. Plenty of sites should start here, get the plant running, and improve later. It is far better than an uncontrolled arrangement nobody has thought about.
A power meter at the incoming supply feeds the inverter's own power control loop, and the inverters curtail to hold a minimum import. It works. Ask two questions before accepting it: what is the loop response time, and what happens if the meter communication is lost. The correct failure behaviour is output to zero, not output to full.
A programmable controller reads generator load, the engine minimum load setpoint and PV output, then writes an active power limit to each inverter over Modbus. This is the arrangement that genuinely saves fuel while protecting the engine. It needs inverters that accept a remote active power setpoint, it needs the register map from the inverter maker, and it needs proper commissioning time rather than an afternoon.
With a battery in the system, surplus PV has somewhere to go and the ramping problem softens considerably. The battery covers the seconds after a cloud passes so the engine never sees the step. It is the technically best answer and the most expensive one, and it only makes sense if the storage is justified on its own terms as well.
Ask for those results in the commissioning file, signed and dated. On a site with a generator, that file is what tells the next engineer why the plant behaves the way it does at ten in the morning on a load-shedding day.