I have this idea: attaching electric motor to clutch and then to a generator then computing when for example we 1kw give the motor how much generator will take back? now we have performance of generator and motor so we can compute clutch performance
Huh? Baring that, as Rory stated, this really doesn't make sense, what do you intend to do?
Would this be a separate test unit outside of the vehicle, or would the electric motor be providing power along with the vehicles gas/diesel engine? Is the transmission still connected to the clutch, driving the wheels?
If it's outside the car (motor drives one clutch plate, other clutch plate drives generator) you still need to account for losses in the motor itself and in the generator. Remember, these are real devices, not theoretical ideal devices, so they will both produce heat lowering your efficiency.
That is, an electric motor drawing 1kW and connected directly to a generator will not produce 1kW of usable power. And the amount of losses will not be constant. At 1kW you might lose, say, 5% due to heat, whereas at 2kW you might lose 15%. You would need to characterize the system before it would even do anything meaningful.
The correct way to test a clutch's performance synthetically would be to use a pulley system and a system of counter weights to apply a frictional torque load to the clutch plate normally connected to the transmission. With a tachometer on both clutch plates, you could find the load that causes the clutch plates to start slipping (engine plate spinning faster than transmission plate). From there I suppose you could calculate the power lost under that particular load, but knowing the load where slipping begins is probably more useful than knowing the power lost. An engaged clutch should never slip under normal operating conditions.
You realise that a clutch, when engaged, will transmit 100% of the power it receives (barring minor friction losses) and when disengaged will transmit 0% (again, barring some minor friction drag)
So performance does not seem like something you will measure here.
You could measure slippage at a particular setting, but again this won't be useful in most cases, as that level will change.