Monitoring Power Levels of a Nuclear Reactor with Seismo-Acoustic Signals Using Machine Learning

Session: Innovative Seismo-Acoustic Applications to Forensics and Novel Monitoring Problems
Type: Oral
Date: 4/29/2020
Time: 04:30 PM
Room: 240
Description: 

The operation of a nuclear reactor involves multiple processes and complex systems of interrelated machines. Variations of the nuclear reactor operational status (power levels) require changes in the operation of these machines, which usually generate seismo-acoustic signals. The seismo-acoustic signals thus carry information on the related changes in operations. Due to the complexity both in the reactor operation and seismo-acoustic wavefield, it is not trivial to decipher operational variables from seismic-acoustic signals. However, machine learning algorithms have shown great potential in extracting useful information from seismic signals for various applications. Using data collected at a three-component seismic station together with acoustic sensors located near the High Flux Isotope Reactor (HFIR) at Oak Ridge National Laboratory (ORNL), we can infer power levels of the reactor directly from seismo-acoustic signals with machine learning techniques.

HFIR is an 85 MW research reactor with operational cycles of about 24 days. After an operation cycle, the reactor is switched to an outage state. When a new cycle starts, the cycle ramp-up consists of step-wise power level increases (e.g. 10%, 30%, 50%, 70% and 90%) before reaching full capacity. Seismo-acoustic signals are used to infer both the state change as well as the step-wise power levels. Our approach consists of two independently trained machine learning models. Model 1 is used to distinguish between operation, transition and outage state. Model 2 distinguishes the step-wise power levels. Preliminary results show combining seismic and acoustic data outperforms individual data types for Model 1. The overall accuracy is over 97% for Model 1 and over 80% for Model 2.

Presenting Author: Chengping Chai

Authors