With the many advances in machine learning in recent years, adopting this technology for accelerator operation offers promising perspectives. At CRYRING@ESR we implemented an operator-grade machine automation application with beam optimization support based on a Genetic Algorithm. This tool was used for optimizing the beam intensity via several machine sub-systems such as injection, ion source, local injector beam line.
One challenge for automatically optimizing the machine is the reliability of the injected beam in terms of pulse-by-pulse intensity variation. To mitigate this beam-variation issue, our recent work aims at developing software discrimination signals to identify „bad“ beam pulses based on time-series data. For this, our team established a rather lightweight framework of tools for on-line data monitoring and mid-term data storage in parallel to the official FAIR Archiving system based on the InfluxDB and Grafana software packages. Besides reporting on the technology stack for this framework we will provide a status update of the development of our „bad pulse“ detection which as a by-product allows user-friendly availability calculation and online monitoring.