EPRI Steam Generator Management Program – O i f R t ......EPRI Steam Generator Management Program...
Transcript of EPRI Steam Generator Management Program – O i f R t ......EPRI Steam Generator Management Program...
EPRI Steam Generator Management P O i f R t NDEProgram – Overview of Recent NDE
Research ActivitiesJames Benson
Technical Executive
5th International CANDU In-Service Inspection Workshop in conjunction with the NDT in Canada 2014 Conferenceconjunction with the NDT in Canada 2014 Conference
June 16-18, 2014
Recent Steam Generator NDE Research ActivitiesActivities
1. Development of Simulation Software for SG Eddy Current InspectionsCurrent Inspections
2. Development of SG Eddy Current Automatic Data Analysis AlgorithmsAnalysis Algorithms
2© 2014 Electric Power Research Institute, Inc. All rights reserved.
D l t f Si l ti S ft fDevelopment of Simulation Software for SG Eddy Current Inspections
Benefits of SG Tube Simulation (SGTSIM) Model• Predict EC probe signals for different SG tube defect geometries• Determine effect of probe wobble, test frequency, sludge
characteristics on probe measurements• Visualization of field/flaw interaction• Optimization of sensor/system design• Optimization of sensor/system design• Test bed for generating defect signatures• Useful in Probability of Detection (POD) models at low costy ( )• Assist in determining root cause of complex signals
K Ad f Si l i M d lKey Advantages of Simulation Model:• Provides an inexpensive and fast method to simulate realistic
defect geometries in steam generator tubing
4© 2014 Electric Power Research Institute, Inc. All rights reserved.
g g g
Practical Applications of Simulator SoftwareSoftware• Utility Engineers− Assist in complex signal interpretation & Tube Integrity Assessments− Assist in POD calculations
• Inspection vendors– Assist in signal interpretationg p– Assist in SSPD development
• Probe developers– Aid in probe design– Aid in probe design
• NDE instructors– Training tool
G t i l f t i i d t– Generate signals for training data• Researchers / Qualifying Institute
– Generate signals for probe technique qualification (ETSS’s)
5© 2014 Electric Power Research Institute, Inc. All rights reserved.
– Generate signals for performance demonstration (QDA/AAPDD)
SG Tube Simulator (SGTSIM) Features -Summaryy
Probes
B bbi X Probe
Tube Geometry
Support Pl
TubeSh
Free SBobbin
Pancake
X-Probe Plate Sheet Span
Defect Geometry+ Point(.115) Defect Geometry
Location Freespan
Orientation Circ
ShapeOther features: Freespan TSP TTS ID/OD
Circ Axial
Circular Elliptical Rectangular Real Cracks
Calibration (Manual/ Auto)
Data Export in MIZ Format
B t h P i Real Cracks Batch Processing
6© 2014 Electric Power Research Institute, Inc. All rights reserved.
Results are validated with experimental eddy current test data
Example of Simulated Signals Exported into Eddynet Environment
Plus Point® Signal from OD Axial Flaw 0.3”L x 40% TW
50 kHz
200 kHz
300100
7© 2014 Electric Power Research Institute, Inc. All rights reserved.
300 kHz
00kHz
Example of Simulated Signals Exported into Eddynet Environment
Plus Point® Signal from OD Axial Flaw 0.3”L x 40% TW
50 kHz 100 kHz 200 kHz 300 kHz
8© 2014 Electric Power Research Institute, Inc. All rights reserved.
50 kHz 100 kHz 200 kHz 300 kHz
Examples of Simulated Signal Exported and Displayed in Eddynetand Displayed in Eddynet
9© 2014 Electric Power Research Institute, Inc. All rights reserved.
Simulated 100% Ax Notch +Point® 200kHz Experimental 100% Ax Notch +Point 200kHz
Example of Exporting Data From a Simulated Flaw SignalFlaw Signal • Simulated data from a single channel can be exported in
Eddynet format y• Simulated data from +Point® and pancake coil probes has
been successfully injected into field data
10© 2014 Electric Power Research Institute, Inc. All rights reserved.
Lissajous & 1D strip plots 3D plot
Example Simulation of a Real Crack – Experimental ValidationExperimental Validation
Crack Profile from MET data
Experimental Signals(Field Data)
+Point ® Probe Data @ 300KHz
MET data
Simulated Signals
11© 2014 Electric Power Research Institute, Inc. All rights reserved.
Vertical Channel Horizontal Channel
Example Simulation of Tube-to-Tube Wear Bobbin Probe simulation compared to actual ECT datap
380 kHz
Length = 10”
Max Depth 0.019” (50% TW)
Simulated Symmetric Wear Depth Profile
Dep
th
Dep
th (m
)
Max Width= 0.194”
∆ W
ear
from
Max
D
12© 2014 Electric Power Research Institute, Inc. All rights reserved.
Wear Length (m)
Example Simulation of Tube-to-Tube Wear Bobbin Probe simulation compared to actual ECT datap
380 KHz
Length = 10”
Depth 0.019” (50% TW)
Maximum Width=0 194”
Simulated Non-Symmetric Wear Depth Profile
Dep
th
epth
(m)
Maximum Width=0.194”
Wear Length (m)
∆ W
ear D
from
Max
De
13© 2014 Electric Power Research Institute, Inc. All rights reserved.
Wear Length (m)
Experimental ValidationArray Coil on X-Probe®: Axial Channel – Vertical – (190 kHz)
Tube OD: 0.628’’; ID: 0.552”; Tube-to-tube wear length 10.1’’; Depth 50%Axial(real) Axial(real)
14© 2014 Electric Power Research Institute, Inc. All rights reserved.
Experimental signals Simulated signals
Development of SG Eddy CurrentDevelopment of SG Eddy Current Automatic Data Analysis Algorithms
Benefits of Automated Data Analysis System
• Faster analysis of large volumes of data compared to manual data analysis.manual data analysis.
• Increase in consistency of degradation detection.
• Increase in accuracy of degradation characterization.
• Addresses concerns about future workforce shortages by g ystoring expert knowledge.
• Eliminate concerns about operator fatigue.Eliminate concerns about operator fatigue.
16© 2014 Electric Power Research Institute, Inc. All rights reserved.
Typical Data Analysis Modules
Preprocessing• Structure detection
Raw Data Calibration • Filtering
• Frequency Mixing• Adaptive Thresholding
ROI Detection
Feature Selection
Classification
17© 2014 Electric Power Research Institute, Inc. All rights reserved.
Classification (Feature Selection, Rule-Base)
Data Analysis Algorithm Development and Validation ProcessValidation Process
Training Data* - w/ Analysis Results
Algorithm DevelopmentAlgorithm Development
Test Data* - Blind Results
* - Data is from EPRI Automatic Analysis Performance
Algorithm Optimization
18© 2014 Electric Power Research Institute, Inc. All rights reserved.
- Data is from EPRI Automatic Analysis Performance Demonstration Database (AAPDD)
Recent Data Analysis Algorithm Development
• Loose part detection algorithms for carbon steel, stainless steel and copper materialsand copper materials
– Rotating Probes
• 115 Pancake Coil
• +Point Coil®
– Array Probes
19© 2014 Electric Power Research Institute, Inc. All rights reserved.
Loose Part Detection ChallengesCase 1: As distance of the loose part from tube wall OD
increases, detection becomes more difficult,
Case 2: As distance of the loose part from TTS decreases,
tubesheet signal becomes dominant and loose part detectiontubesheet signal becomes dominant and loose part detection
becomes more difficult.
Loose Part
dist_tubeOD
Loose Partdist_TTS
20© 2014 Electric Power Research Institute, Inc. All rights reserved.
Case 1 Case 2
D l t f Al ith tDevelopment of Algorithms to Automatically Detect Loose Parts
From X Probe® Array DataFrom X-Probe® Array Data
Carbon Steel Loose Parts
21© 2014 Electric Power Research Institute, Inc. All rights reserved.
Loose Part Test Configurations
a: Distance from Tube Wall ODb: Distance above TTS
a: 0-5mm
b: 0-18mmFor laboratory mock-up:
Loose Part
Ste
elLoose parts used in laboratory experiments
Car
bon
S
22© 2014 Electric Power Research Institute, Inc. All rights reserved.
Spacers used in laboratory experiments
Calibration & Phase Correction
File: LB00HCAL00702/DHR000C009I014 Circ. Channel 50 KHz, Vert. component, 18mm above TTS, in contact with tube
Before Calibration After CalibrationBefore Calibration After Calibration
20 2020
40
60
80
20
40
60
80
Tube
100
120
140
100
120
140Tube sheet
Loose Part
2 4 6 8 10 12 14 16 18
160
180
2005 10 15 20 25 30 35
160
180
200
23© 2014 Electric Power Research Institute, Inc. All rights reserved.
Step 1-TTS Detection & Segmentation File: LB00HCAL00702/DHR000C009I007 50 KHz, Vert. component,
2mm above TTS, in contact with tube
Axial projection P(i)
Derivative of P, D(i) = P(i) - P(i-1)
Location of detected TTS by thresholding D(i)
0 0 20 40 60 80
100
120 0-10 -5 0 5 10 15 20
P(i) D(i) = P(i)2D data
20
40
60
Tube
2040
60
2040
60
Loose Part
80
100
120
80100
120
80100
120
140
160
180
Loose Part
Tubesheet0
140160
180
0140
160180
24© 2014 Electric Power Research Institute, Inc. All rights reserved.
5 10 15 20 25 30 35200
0200
0200
Step 2-Preprocessing - Median RemovalFile: LB00HCAL00702/DHR000C009I014 50 KHz AF, Vert. component, 18mm above TTS, in contact with tube
Before median removal
Segmented data After median removal
10
20
10
20
Input for step 3
30
40
30
40
L P t50
60
70
50
60
70
Loose Part
70
80
90
70
80
90
25© 2014 Electric Power Research Institute, Inc. All rights reserved.
5 10 15 20 25 30 35
5 10 15 20 25 30 35
Step 3- Adaptive Thresholding and ROI Detection File: LB00HCAL00702/DHR000C009I025 50 KHz AF, Vert. component,
S t d I Th h ldi d ROI d t ti
10
Segmented Image Thresholding and ROI detection
20
30
40 Loose Part L P t40
50
60
Loose Part Loose Part
5 10 15 20 25 30 35
70
Residue from TTS
26© 2014 Electric Power Research Institute, Inc. All rights reserved.
Residue from TTS
18mm above TTS, in contact with tube
Step 4- Feature Extraction & Classification
•Features extracted from calibrated data inside ROI boxes:– Maximum and minimum signal amplitude g p
– The corresponding phase at maximum and minimum signal amplitudesignal amplitude
•Rule-based classification applied to the extracted featuresfeatures
27© 2014 Electric Power Research Institute, Inc. All rights reserved.
Step 4- Feature Extraction & ClassificationFile: LB00HCAL00702/DHR000C009I025 50 KHz AF, Vert. component,
18 b TTS i i h b18mm above TTS, in contact with tube File: LB00HCAL00702/DHR000C009I025 50 KHz AF, Vert. component,
10
Segmented Image Thresholding and ROI detectionClassification
10
20
30
L P t40
50
60
Loose Part Loose Part Loose Part
5 10 15 20 25 30 35
70
28© 2014 Electric Power Research Institute, Inc. All rights reserved.
False ROI: Residue from TTS
Training Data Distribution & Result Type/Total tubes
# of tubes in training set
LP#1 / 10 6
• 100 files were randomly chosen and used for training• The set contains both lab data (LP#1-10) and field data (LP#11 & 12)• AA was trained based on this 100 flies
LP#2 / 10 2
LP#3 / 10 3
LP#4 / 50 14
LP#5 / 50 9
LP#6 / 10 1
LP#7 / 10 1
LP#8 / 10 1
LP#9 / 51 16
LP#1-LP#10: *LP#11: Carbon steel weld backing ring *LP#12: Carbon steel nail
LP#10 / 28 13
*LP#11 / 51 14
*LP#12 / 51 17
# tubes analyzed
Detection rate by human analyst
(PLP+WLP)
Detection rateby AA
(PLP+WLP)
NDD by human analyst
NDD by AA AA False Call Rate
/Tube
100 79% 85% 21% 15% 0.14(14/100)
"PLP" - Possible Loose Part (High confidence)"WLP" Weak Loose Part signal (Not likely to be reported in a field exam)
29© 2014 Electric Power Research Institute, Inc. All rights reserved.
WLP - Weak Loose Part signal (Not likely to be reported in a field exam)"NDD" - No Loose Part signal - NDE technique limitation
Blind Test data
• After training was performed on the remaining 269 tubes
# tubes analyzed
Detection rate by human analyst
(PLP+WLP)
Detection rateby AA
(PLP+WLP)
NDD by human analyst
NDD by AA AA False Call Rate
/Tube
• "PLP" Possible Loose Part (High confidence)
269 71% 76% 29% 24% 0.21(58/269)
• "PLP" - Possible Loose Part (High confidence)• "WLP" - Weak Loose Part signal (Not likely to be reported in a field exam)• "NDD" - No Loose Part signal - NDE technique limitation
• Analysis of other loose part materials (e.g. stainless steel and copper) is in progress
30© 2014 Electric Power Research Institute, Inc. All rights reserved.
and copper) is in progress
Summary
• EPRI SGMP conducts research to improve SG NDE capabilities, including:
detection sizing and characterization of tube degradation– detection, sizing and characterization of tube degradation– detection of foreign objects
• Results from SG NDE are used as input to decisions on:– tube repairs, inspection intervals, repair methods, deposit
removal processes and the need for foreign object retrievalretrieval
• Important aspects of the SGMP NDE research include:– development of eddy current simulation modeling tools– development of efficiency improvements of data analysis
through automation – qualification of data analysis processes and systems
31© 2014 Electric Power Research Institute, Inc. All rights reserved.
qualification of data analysis processes and systems
Together…Shaping the Future of Electricity