Investigation of gas circulator response to load transients in nulcear power plant operation

Gas circulator units are a critical component of the Advanced Gas-cooled Reactor (AGR), one of the nuclear power plant (NPP) designs in current use within the UK. The condition monitoring of these assets is central to the safe and economic operation of the AGRs and is achieved through analysis of vibration data. Due to the dynamic nature of reactor operation, each plant item is subject to a variety of system transients of which engineers are required to identify and reason about with regards to asset health. A characteristic feature of AGR is the ability to refuel during reduced generation, known as 'low power refueling' (LPR). The LPR event leads to a change in operational state for the gas circulators, with the vibration profile of each unit reacting accordingly to the transient conditions. It has been suggested that the changing conditions subject to these items during LPR and other such transients may cause damage to the assets. From these assumptions, it is proposed that useful information on gas circulator condition can be determined from the analysis of vibration response to the LPR event. This paper presents an investigation into asset vibration during an LPR. A machine learning classification approach is used in order define each transient instance and its behavioral features statistically. Classification and reasoning about the regular transients such as the LPR represents the primary stage in modeling higher complexity events for advanced event-driven diagnostics, a move from the previous focus on simplistic alarm boundary limits.