Sealab

Facility for testing a superconducting radio frequency electron gun

Offsets for input parameters 0-14 and output parameters 15-18. Blue dots: Randomly added generated offsets. Magenta stars: Predicted offsets by optimization algorithm.

A photoelectron gun cavity is an electron source that every Energy Recovery Linac (ERL) requires. Such a device is also installed in Sealab (formerly known as BerlinPro), a test facility with the aim to build a fully superconducting photoelectron injector. However, the inital setup and calibration requires a lot of time and experience. We have successfully applied parameter determination and beamline optimization in simulation environments. Parameter determination is the identification of offsets between the simulation and the real-world device. For this, we have build a surrogate model using Neural Networks, that means a fast approximation of the complex simulation. For optimization we used common global optimization algorithms and an advanced Deep Reinforcement Learning approach.