SHM CVEN 631

what is SHM

provide updated insights on structural performance under aging and environmental effects by identifying if damage is present and how fast will the damage grow and exceed a critical level

SHM key process

observation of structure over time through dynamic response measurements

extraction of damage-sensitive features

stat analysis of features to assess system health

economic benefits SHM

accelerated reoccupation of manufacturing facilities reduces economic losses after seismic events

run-to-failure maintenance

operate system until critical component failure

no monitoring system investment required

high costs and unacceptable risk for life-safety

time-based maintenance

replace or service components at predefined intervals

doesn’t account for actual condition

maintenance approach requires that critical components are serviced or replaced at predefined time or use regardless of condition

condition-based maintenance (via SHM)

utilizes sensing systems to monitor structure responses

alerts operators to damage or degradation for timely action

define damage for SHM

international/unintentional changes to material or geometric of system including changes to boundary conditions and system connectivity, which affect current or future performance of system

changes structure’s mass, stiffness, or energy dissipation and can impact boundary conditionsto the system's integrity and functionality

Statistical pattern recognition

1. operational evaluation

2. data acquisition

3. feature selection

4. statistical modeling for feature discrimination

role of machine learning

applied in feature selection and statistical modeling

goal: learn the relationship bw features derived from data and damaged state of structure

supervised learning

training data is labeled with multiple classes; can do group classification and regression analysis

unsupervised learning

no class labels; identifies intrinsic relationship within the data (what changed from normal)

can construct single-class models for novelty detection; only when undamaged data available; focus on outlier detection or novelty detection

model-based SHM

builds physics-based (hard to be accurate) or law-based model (law-based easier)

adjust model to align w real world data and relies on linear algebraic method

monitors deviation from normal conditions and updates based on location and extent of damage

requires training data for initial model updates

Challenge: difficulty in modeling specific joints and material properties

Data-driven SHM

relies on training data from healthy and damaged states and uses pattern recognition

builds statistical models and incorporates machine learning

integration w law-based models

focuses on statistical modeling instead of physics-based modeling

Operational Evaluation

research on related infrastructure

purpose:

set boundaries on what will be monitored and how monitoring will be executed

tailors damage identification process to unique features of system and damage being detected

SHM customizable and one may not work for another

Data Acquisition

Key Component:

selection of excitation methods

sensor types, quantities, and locations

data acquisiton, storage, and transmission hardware

Application-specific considerations:

1. economic factors influence hardware decision

2. data collection intervals depend on application

Data Normalization

def: separates changes in system response due to operational/environmental variability from those caused by damage

ex. tempearture effect, traffic variation

Data Cleansing

filter/check if want to accept or reject data

based on knowledge gained by indv directly involved w data acquisition

signal processing techniques

Data Compression

process of reducing the dimension of measured data for easier management

data fusion

process of combining information from multiple sources to enhance fidelity of damage detection process

combines data from spatially distributed sensors

heterogenous data types include kinematic response, envrionment parameters, and measures operational parameters to determine if damage present

related to data normalization cleaning and compression processes

Feature Selection

quantities extracted from system response data indicating presence of damage

ideal: low dimensional and highly sensitive to structural conditions

enhancement in feature selection: fuse data from multiple sensors and employ data normalization

statistical modeling for feature discrimination

goal: develop algorithms to quantify damage state of structure using extracted features

challenge: difficult to define relationship bw features and damage using physic based method

ML tech employed

obj: aim to reduce false diagnoses

economic and life safety justification

clear and measurable benefits

economic benefits: decrease maintenance cycle and warranty and increase manuf capacity

life safety benefits: hard to justify

5 hierarchical levels of damage

1. type of damage

2. threshold level

3. critical level

4. locations

5. rate of growth

three challenges of placing SHM sensors for monitoring

1. sensor placement limitation; can’t interfere with traffic flow

2. size of specimen; larger = harder

3. environment condition; weather and moisture can effect data reading if not taken to account during design phase