Detecting Threatening States in Laser Beams

This poster presents our approach to automatic detection of critical failure states in the pulsed Petawatt laser systems DRACO and PENELOPE, used for investigations of exotic states of matter and medical applications. The beam shape is controlled to avoid high destructive energy densities. However, randomly occurring states threatening the device must be detected between pulses and trigger an interlock in the device firing at 10Hz.

The states we are aiming to detect are rare; thus, training data for this category is scarce. To address this, we present two approaches: First, to identify regions of interest based on physical properties of the system and apply a convolutional neural network (CNN) to to identify true positives. Secondly, using CNN-based image segmentation to localize and classify regions of interest.