Classification of Industrial Facility Power-Levels Using Seismo-Acoustic Signatures

Industrial operations may be broadly attributed to the initiation-or-termination of machine processes and changes in operational state configurations (e.g. power consumption, speed level, and frequency). These operational activities, within or adjacent to industrial facilities, generate mechanical energy that may propagate into the earth and air as seismic and acoustic waves, respectively. The types and intensities of the seismo-acoustic signals vary according to machinery and their relative location with respect to sensors. We have analyzed continuous seismo-acoustic data from a permanent station nearby the High Flux Isotope Reactor (HFIR) at Oak Ridge National Laboratory (ORNL). HFIR is an 85 MW research reactor that runs on an operational cycle of about 24 days, followed by an outage period. For each cycle, the reactor starts with a set of increasing power-levels (0%, 10%, 30%, 50%, 70%, and 90%) before reaching full 100% capacity. The recorded data provide us an opportunity to monitor power-levels using sensors outside the facility. Seismo-acoustic data corresponding to multiple cycle start-ups are extracted and compared to facility-operational ground-truth information. We will present results on the comparison of single and multiple seismic channels, as well as multimodal (seismic and acoustic) automatic classification of reactor start-up power-levels obtained by training boosted learning models.