8/2/2019 SHM_Intro
1/13
LA-UR-07-3231 Unclassified
Introduction to Structural Health Monitoring andFeature Extraction
Charles R. Farrar
Presented at
Engineering Institute Workshop
July 25th, 2006
Engineering Institute
8/2/2019 SHM_Intro
2/13
LA-UR-07-3231 Unclassified
Structural Health Monitoring
Engineering Institute
8/2/2019 SHM_Intro
3/13
LA-UR-07-3231 Unclassified
Motivation for Structural Health Monitoring
Move from time-based maintenance to condition-based maintenance
Combat asset readiness
ew us ness mo e s
Manufacturers of large capital investment hardware can charge by theamount of life used instead of a time-based lease.
-with need to integrate diverse technologies makes StructuralHealth Monitoring a Grand Challenge problem for aerospace,
Engineering Institute
c v an mec an ca eng neers n e cen ury
8/2/2019 SHM_Intro
4/13
LA-UR-07-3231 Unclassified
Definition of Damage Damage is defined as changes to
the material and/or geometric
adversely affect its performance.
All materials used in engineeringsys ems ave some n ereninitial flaws.
Under environmental and
grow and coalesce to producecomponent level failure.
-level failure.
Must consider the length andtime scales associated with
Engineering Institute
amage evo ut on.
8/2/2019 SHM_Intro
5/13
LA-UR-07-3231 Unclassified
Classifying Damage Identification Methods
Damage Identification
Microscopic flaw/damage identification Used to develop material failure models
Incipient, macroscopic, material/component level damage Non-destructive evaluation (local, off-line inspection)
- - ,on-line)
Component damage/failure system level damage Structural health monitoring Condition monitoring (applied to rotating machinery) Health and usage monitoring systems (HUMS, Rotor craft) Statistical process control (monitors system processes where damage
can be one cause of loss of rocess control
Damage Prognosis Adds prediction of remaining life capability to SHM
Engineering Institute
8/2/2019 SHM_Intro
6/13
LA-UR-07-3231 Unclassified
Structural Monitoring SHM
20
21
R
183 6
57 10
Ch. 2
Ch. 10Ch. 3
15
16
17
18
19Roof
Ch. 4
Ch. 11Ch. 5
N
9
10
11
12
14
177 3
16th Floor
Ch. 6
Ch. 12 Ch. 7
3
4
5
6
7
8
3RD Floor
9th Floor
Ch. 8
Ch. 13Ch. 9
Structural monitoring is not
G
B
2
Basement
Ch. 1used to make decisionsregarding system operationand maintenance
Engineering Institute
8/2/2019 SHM_Intro
7/13
8/2/2019 SHM_Intro
8/13
LA-UR-07-3231 Unclassified
The Structural Health Monitoring Process
1. Operational evaluationDefines the damage to be detected and begins to
issues for a structural health monitoring system.2. Data acquisition
used in the feature extraction process.
3. Feature extraction -
Data Cleansing
Data Normalization
information from measured data.
4. Statistical model development for
Data Fusion
Data Compression
(implemented by
feature discriminationClassifies feature distributions into damaged or
undamaged category.
so ware an orhardware)
Engineering Institute
8/2/2019 SHM_Intro
9/13
LA-UR-07-3231 Unclassified
Introduction to Features What is a Feature?
A feature is some characteristic of the measured response that is,
other signal inspection technique Feature extraction transforms data into information It is desirable to have exam les of the features from both
damaged and undamaged structures Primary Characteristics of features
Sensitivity - Feature should ideally be very sensitive to damage
Dimensionality - Want the feature to have the lowest dimensionpossible
Com utational Re uirements - Features should be com utable
with minimal assumptions and CPU cycles Want to use the simplest feature possible that can
distinguish between the damaged and undamaged
Engineering Institute
sys em
8/2/2019 SHM_Intro
10/13
LA-UR-07-3231 Unclassified
Feature Extraction
Approaches to identifying damage-sensitive features. as exper ence
Component and system testing Numerical analysis to simulate damaged system response
Features types. Absolute (derived from single data source, e.g. modal
frequency) Relative (derived from multiple data sources, e.g. mode shape)
Damage sensitive features fall into three categories. Waveform or image comparison
Model parameters
Residual errors between measured and predicted response.
Engineering Institute
8/2/2019 SHM_Intro
11/13
LA-UR-07-3231 Unclassified
Feature Extraction Want many samples of low-dimension feature vectors. The need for low-dimension feature vectors often
(e.g. compression of accel.-time histories into modalproperties).
multiple and possibly heterogeneous sources
(estimation mode shapes). .
Ideally, the features should change monotonically withdama e level.
Identify and quantify sources of feature variability.
Incorporate feedback from data acquisition and
Engineering Institute
.
8/2/2019 SHM_Intro
12/13
LA-UR-07-3231 Unclassified
Data Normalization
1000
500
0
500
1000
Strain
Signal 1
Data Point # = 26980Time Period = 601.17 sec t = 0.02228 sec
0 100 200 300 400 500 600 700
1000
500
0
500
1000
Strain
Signal 2
7.650 100 200 300 400 500 600 7001000
500
0
500
1000
Strain
Signal 3
14.7
20.35
7.45
18.3
7.35
7.50
.
uency(Hz)
Mean Frequency Value
(Top West Outdoor - Top East Outdoor )
Time (Sec)
1.1
-5.1
-10.5
1.9
-0.2 0.6 0
3
7.05
7.20
FirstFreq
Engineering Institute
9:15 11:30 13:12 15:13 17:52 20:09 21:20 23:29 1:21 3:19 5:19 7:03 9:22
8/2/2019 SHM_Intro
13/13
LA-UR-07-3231 Unclassified
Features are Used to Answer the Following:1 Is the system damaged?
Group classification problem for supervised learning Identification of outliers for unsu ervised learnin
2 Where is the damage located? Group classification or regression analysis problem for supervised
learning Identification of outliers for unsupervised learning
3 What type of damage is present?
Can only be answered in a supervised learning mode roup c ass ca on
4 What is the extent of damage? Can only be answered in a supervised learning mode
roup c ass ca on or regress on ana ys s5 What is the remaining useful life of the structure? (Prognosis)
Can only be answered in a supervised learning mode
Engineering Institute
egress on ana ys s