Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
1
Uncertainty Characterization of the EPRI/NRC
Weld Residual Stress Round Robin Data
Michael L. Benson and Matthew J. HomiackU.S. Nuclear Regulatory Commission
2016 International LWR Materials Reliability Conference and Exhibition
August 1-4, 2016
Chicago, IL
The view expressed herein are those of the authors and do not reflect
the views of the U.S. Nuclear Commission
2
Background and Introduction
3
WRS Research at NRC
• 4-year program under MOU with EPRI, ended in
December 2015
• Number of measurement and modeling studies– Small, scientific specimens
– Full-scale mockups and components from cancelled
plants
• NUREG-2162 and MRP-316
•Scientific Weld Specimens•Phase 1A: Restrained Plates (QTY 4)
•Phase 1B: Small Cylinders (QTY 4)
•Purpose: Develop FE models.
Ph
ase
1 -
EP
RI
•Fabricated Prototypic Nozzles
•Type 8 Surge Nozzles (QTY 2)
•Purpose: Prototypic scale under controlled conditions. Validate FE models.
Ph
ase
2 -
NR
C
•Plant Components•WNP-3 S&R PZR Nozzles (QTY 3)
•Purpose: Validate FE models.
Ph
ase
3 -
EP
RI
•Plant Components•WNP-3 CL Nozzle (QTY 1)
•RS Measurements funded by NRC
•Purpose: Effect of overlay on ID.
Ph
ase
4 -
EP
RI
4
Phase 2a Study
• Reasonable agreement on average between
measurements and models
• Significant scatter about the average
• Incrementally providing participants more information
(e.g., material properties, thermocouple
measurements) did not improve the scatter
• Gained confidence in the hole drilling and contour
measurement techniques
• Future work: improving, characterizing,
understanding the prediction uncertainty
5
Phase 2b Study
• Second pressurizer surge line mockup, similar to Phase 2a
• Hole drilling and contour measurements
• Crafted a set of modeling guidelines
• Participant questionnaire
• Initiated a second international round robin
• Obtained another dataset of measurements and models
6
Sandia National Laboratory Uncertainty Project
• NRC recognized a need for more sophisticated
treatment of the round robin dataset
– Quantify uncertainty
– Discuss measurement-model comparisons
• Sandia has worked on NRC/EPRI’s Extremely Low
Probability of Rupture (xLPR) project, with expertise in
statistics and probability
• Sandia recently finished a project to describe
uncertainty in the Phase 2b dataset
• Objective: overview and introduction of Sandia’s work
7
Overview of Phase 2b Results
8
Hole Drilling Measurements
• Four 1.5-mm diameter holes drilled through the
centerline of the dissimilar metal weld, 90° apart
• One-dimensional representation of the stress variation
through the wall thickness
• Both axial and hoop stresses determined from a single
measurement
• Natural representation is from OD to ID, given that
drilling procedure starts on OD
9
Contour Measurements
• Contour measurement requires completely destroying
the part
• Two-dimensional representation of the stress state on
the relevant cut planes
• Axial stresses measured at centerline of the dissimilar
metal weld between the nozzle/butter and the safe end
• Hoop stresses measured along the axial position of the
part
• One-dimensional stress profiles can be extracted from
this data
mid-weld path
10
Modeling Results
• Results from two hardening laws: isotropic and
kinematic
• Two-dimensional contour plots provided by participants,
but no raw data
• Raw data supplied as one-dimensional profile along the
dissimilar metal weld centerline, reported ID to OD
Isotropic Hardening
Kinematic Hardening
11
Need for Data Processing
• Ultimate objective is to benchmark finite element results against the measurements; not straight-forward
• Various datasets require processing the raw forms before making meaningful comparisons
– Sort
– Interpolate
– Normalize
– Convert 2-D data to 1-D data
• Various measurements
– What’s the correct benchmark?
– Simple average?
– Use of multiple benchmarks?
– What’s the uncertainty of benchmark(s)?
12
Overview of Sandia’s Methodology
13
Describing Functional Data
• WRS is functional data
• Two aspects of functional data
– Amplitude variability
– Phase variability
• In the xLPR project, a method to describe uncertainty and statistically sample WRS was developed
– Kurth and Sallaberry, et. al: PVP2016-63962 and PVP2016-63963
– Accounts for expected features of a WRS curve: point-to-point correlation, force balance requirements
– Harmonize the two methods in the future?
dfWRS
phase
amplitude
14
Step 1: Smooth Data
• WRS modeling data is discrete by nature, but WRS is a
continuous function of depth in reality
• Smoothing with spline functions allows for representation of
the WRS as a continuous function
• Smoothing can introduce additional uncertainty, which must
be assessed
Raw Data
Smoothed Data
15
Step 2: Register the Data
• Aligns the data
• Warping functions
• Characterizes and removes phase variability
Smoothed Data
Registered Data
16
Step 3: Construct Probabilistic Model
• The probabilistic model is based upon Functional Principle
Components Analysis
• Modeling amplitude variability requires registered (aligned) data
• Modeling phase variability requires the warping functions used to
register the data
• Combine the two models, allowing sampling
17
Step 4: Bootstrapping
• A statistical sampling method used to estimate uncertainty on a distribution parameter
– mean
– quantiles
• Iteratively sample from the probability model, estimating the parameter (e.g., the mean) each time
• Construct confidence bounds based upon the results, without a priori assigning a distribution type to the data
• Now, we can estimate how well we know the mean and confidence bounds
18
Closing
19
Summary
• Provided broad overview of a method to characterize uncertainty in the Phase 2b round robin dataset
• Discussed only the method for characterizing modeling uncertainty
• Characterizing measurement uncertainty also challenging
– WRS not measured directly, so the “measurement” is actually a combination of measurement and models
– Limited data
– Diverse data: axial stress vs. hoop stress, contour method vs. hole drilling techniques
• Comparing measurements to models
– What constitutes a “reasonable” prediction?
– What hardening law provides the best prediction?