Session: 11-01: Nuclear Power NDE

Paper Number: 133645

133645 - Spread Spectrum Time Domain Reflectometry (SSTDR) and Frequency Domain Reflectometry (FDR) Cable Inspection Using Machine Learning 

Abstract: 

 Cables are initially qualified for nuclear power plant use for 40 years. As plants extend their operating license to 60 and 80 years, justification for continued cable use must shift to a performance-based approach since it is cost prohibitive to completely replace cables that are likely still capable of performing their design function. A variety of cable tests are available and are commonly applied during outages when the cables can be taken out of service. Reflectometry tests however use a low voltage high frequency chirp that may be applied for online testing.  Reflectometry data can detect changes in the cable insulation condition that can indicate thermal, radiation, or mechanical damage, can identify splices, phase-phase low resistance faults, water exposure, and other cable anomalies.  Data responses are complex and may be difficult to identify subtle indications in the data.  If these tests are applied online, advances in machine learning may be applied to improve signal diagnostics and call attention to early damage indications such that cable management mitigation steps can be applied to manage further inspections and planned repair or replacement. 

Pacific Northwest National Laboratory’s (PNNL)’s Accelerated and Real Time Experimental Nodal Analysis (ARENA) cable motor test bed was used to test a commercial spread spectrum time domain reflectometry (SSTDR) system, a laboratory instrument software controlled SSTDR, and a vector network analyzer based frequency domain reflectometry (FDR) system’s response to various cable anomalies.  The three instrument systems were able to interrogate cables over a range of frequency bandwidths that can be helpful for human data analysis. Data were presented to both supervised and unsupervised machine learning (ML) analysis to distinguish normal undamaged cable responses from anomalous cable responses.  In an initial test that included approximately 500 data samples, both supervised and unsupervised  ML approaches produced encouraging results with an undamaged/anomalous prediction accuracy from 0.69% to 0.87%.  Recommendations were for an increased and more balanced sample set particularly including more training data.  This paper reviews results of the initial test and preliminary results of an extended test program. 

Presenting Author: Samuel Glass Pacific Northwest National Lab

Presenting Author Biography: S.W. (Bill) Glass has focused his 40+ year career on inspection and robotic technologies mostly related to nuclear power plant operation and decommissioning. Following a BS in Mechanical Engineering and an MS in Bio Engineering from NC State University, he worked for 3 years with the Swedish equivalent of NIOSH. He then joined AREVA working in Virginia and France. He served in numerous engineering and management positions rising to the level of NDE Technical Center Director and expert fellow in NDE and Robotics. In 2015, he joined Pacific Northwest National Laboratory (PNNL) as a technical advisor continuing his interest in inspection and robotic technologies. He has authored more than 100 technical and scientific papers, holds 8 patents and is a Licensed Professional Engineer.

Authors:

S. W. Glass Pacific Northwest National Lab
J. R. Tedeschi Pacific Northwest National Laboratory
M. P. Spencer Pacific Northwest National Laboratory
J. Son Pacific Northwest National Laboratory
M. F. N. Taufique Pacific Northwest National Laboratory
D. Li Pacific Northwest National Laboratory
M. Elen Pacific Northwest National Laboratory
L. S. Fifield Pacific Northwest National Laboratory
J. A. Farber Idaho National Laboratory
A. Al Rashdan Idaho National Laboratory