TY - JOUR
T1 - Development of the Machine Learning-based Safety Significant Factor Inference Model for Diagnosis in Autonomous Control System
AU - Lee, Joomyung
AU - Lin, Linyu
AU - Athe, Paridhi
AU - Dinh, Nam
N1 - Funding Information:
This work is supported by the Advanced Research Projects Agency-Energy (ARPA-E) MEITNER program under the project DE-AR0000976 entitled: “Development of a Nearly Autonomous Management and Control System for Advanced Reactors.” The authors would like to thank Dr. Jeffrey Lane of Zachry Nuclear Engineering for his inputs and support, and Dr. Pascal Rouxelin of North Carolina State University and Dr. Paolo Balestra of Idaho National Laboratory for their support in data generation and processing.
Publisher Copyright:
© 2021 Elsevier Ltd
PY - 2021/11
Y1 - 2021/11
N2 - As a critical component to the autonomous control system, Digital Twin for Diagnosis (DT-D) is a virtual replica of physical systems for an accurate understanding of reactor states. Since the physical damage state cannot be measured directly in transient or accident conditions, safety significant factor (SSF) is introduced as a surrogate index for physical damage states to support safety-related decision making. This study develops a machine learning (ML) based SSF inference model (SSFIM) using the Recurrent Neural Network (RNN) with acceptable accuracy, generalization capability, effectiveness, and robustness against sensor errors. To demonstrate the capability of the ML-based SSFIM, case studies are implemented on a plant simulator for Experimental Breeder Reactor – II. For partial loss of flow accident scenarios, the SSFIM is able to infer the peak fuel centerline temperature with minimally one sensor. Meanwhile the SSFIM is also found to be robust against manipulated sensor drifts and/or random noises.
AB - As a critical component to the autonomous control system, Digital Twin for Diagnosis (DT-D) is a virtual replica of physical systems for an accurate understanding of reactor states. Since the physical damage state cannot be measured directly in transient or accident conditions, safety significant factor (SSF) is introduced as a surrogate index for physical damage states to support safety-related decision making. This study develops a machine learning (ML) based SSF inference model (SSFIM) using the Recurrent Neural Network (RNN) with acceptable accuracy, generalization capability, effectiveness, and robustness against sensor errors. To demonstrate the capability of the ML-based SSFIM, case studies are implemented on a plant simulator for Experimental Breeder Reactor – II. For partial loss of flow accident scenarios, the SSFIM is able to infer the peak fuel centerline temperature with minimally one sensor. Meanwhile the SSFIM is also found to be robust against manipulated sensor drifts and/or random noises.
KW - Diagnosis
KW - Digital twin
KW - Machine Learning
KW - Recurrent Neural Network
KW - Safety significant factor
UR - https://www.scopus.com/pages/publications/85109449538
UR - https://www.mendeley.com/catalogue/c75cf703-61a9-3a7a-843b-8d2dd6bdb19c/
U2 - 10.1016/j.anucene.2021.108443
DO - 10.1016/j.anucene.2021.108443
M3 - Article
AN - SCOPUS:85109449538
SN - 0306-4549
VL - 162
JO - Annals of Nuclear Energy
JF - Annals of Nuclear Energy
M1 - 108443
ER -