Year
2019
Abstract
Changes in the status (interruption or initiation) or configuration (change in power, speed, or frequency) of machines in industrial environments are considered operational events. These events are characterized by transient or permanent changes in the spectral distribution of seismic and acoustic energy that can be observed in the local wavefield. We developed an approach for the automatic detection of operational events based on identifying statistical changes in the spectral distribution of energy compared to a spectral baseline. We tested our detection framework using data from a seismoacoustic station located inside the facility fence of a small nuclear research reactor (High Flux Isotope Reactor at Oak Ridge National Laboratory, Tennessee, US) and built an event catalog that covers a full cycle of the reactor. The detections in the event catalog are validated against the reactor’s ground-truth information. The catalog shows an increased number of detections as the reactor is in shutdown mode, which may be related to system testing before a startup. A recurrent event is observed in the catalog with detections only in the seismic channel. This event displays short-time (around 10 minutes), small amplitude, and a rapid energy onset and correlates well with the operations of letdown valves from the secondary coolant system.