Efficiency improvement of pulse waveform shaping on high power laser facility using deep learning

Abstract

Laser pulse shaping is one of the key and time-consuming processes for preparing a shot on a high power laser facility. The frequent shifts of laser pulse waveform between shots makes improving laser pulse shaping efficiency an urgency. By combining historical dataset accumulated by iterative way of pulse shaping system and prevailing deep learning method, a U-Net modal is trained and applied on our laser facility. Now the pulse waveform shaping system integrated with this modal is capable of shaping and qualifying most of laser pulse waveforms in less than 5 s. As for part of unqualified outputs, a strategy of first deep learning prediction and then iterative way is able to fix all the rest, which cuts down roughly 80 % percent of time consumption, comparing with the absolutely iterative way.