Improvements of expert system for RF-power stations

Superconducting linear accelerators are becoming more and more complex. Despite of the growth of their scale they have to meet very demanding norms concerning availability and reliability. Taking into account these facts, high degree of automation is obligatory for several accelerator subsystems. This article aims at summing up the effort of improvement of expert system for RF-power station for the FLASH. The main purpose of this software is to facilitate operators with automatic driving the power supply subsystem. Owing to the high level of autonomy possessed by the software special care has to be taken to assure predictability and safety of AI-aided operation. To assure sound engineering foundations of the project, techniques that facilitate the reasoning about correctness and predictability its behaviour have been used. As a starting point of design finite state machine computational model has been chosen. After a couple of months of lighting with several approaches of the FSM design and facing state explosion problem new design emerged which is expected to be a lot easier to maintain and adapt to new requirements