Automation with Bayesian optimization in the Karabo SCADA system

Automation within the control system environment is a strategic goal at the European XFEL facility for various reasons: staff can be allocated more efficiently, procedures can be standardized to increase data quality, operator errors are minimized, and less experienced users can operate instruments. Prime candidates for automation are often recurring procedures during facility operation, such as beam-alignment tasks. Therein, a set of motors/mirrors is moved to optimize a characteristic property of the system at hand, such as the beam intensity or the beam position. This type of problem can be described more generally as the optimization of an expensive-to-evaluate, and often multivariate black-box function. A well-established and efficient method to address such problems is Bayesian Optimization (BO). In this contribution a software is presented, which uses BO to automate the setup of scientific instrumentation, and which is highly adaptable towards a broad range of use-cases. The software is implemented within the Karabo supervisory control and data acquisition system and uses the botorch library for BO.