aircraft component structural health · pdf filecopyright © twi ltd 2012 ultrasonic...
TRANSCRIPT
Copyright TWI Ltd 2012
Aircraft ComponentStructural Health Monitoring
The SelfScan Project
14 March 2012 Alex Haig
Copyright TWI Ltd 2012
SelfScan PartnersBeneficiary name Country Type
1 TWI Ltd (Coordinator) UK RTD2 Optel Poland SME3 Phillips Consultants UK SME4 Isotest Engineering s.r.l Italy SME5 Smart Material GmbH Germany SME6 Cereteth Greece RTD7 NDT Expert France LE
Research for the benefitof SMEs
SELFSCAN project has received funding from the European Unions Seventh Framework Programme managed by REA - Research Executive Agency http://ec.europa.eu/research/rea [FP7/2007-2013] under grant agreement number: 232212
Copyright TWI Ltd 2012
Project Goals
Novel structural health monitoring system
Light, flexible transducers
High defect detection sensitivity
Automated in-service defect detection
Copyright TWI Ltd 2012
Contents The approach:
Medium sized, inaccessible, complex parts Ultrasonic Guided Waves
Overview of development process Understanding the ultrasonics Building the systems Gathering the data
Neural Networks Development Training Testing
Copyright TWI Ltd 2012
Ultrasonic Guided Waves Background experience
Mainly oil and gas Thick steel 20 to 80 kHz
Guided waves are good for Long range (~tens of meters) Volumetric defect sensitivity Monitoring inaccessible components
Copyright TWI Ltd 2012
Aircraft Components Identified two main classes of critical
structure component Aluminium skin panel
Complex due to rivets and layers Approximately 2 mm to 7 mm thick
Load bearing components 10 to 20 mm thick Aluminium or steel Complex shape
Copyright TWI Ltd 2012
Aircraft Components Two main classes of critical structure
component Aluminium skin panel
Complex due to rivets and layers Approximately 2 mm to 5 mm thick
Load bearing components 10 to 20 mm thick Aluminium or steel Complex shape
Do not requirelong range
Small defectsare significant
Copyright TWI Ltd 2012
Aircraft Components Two main classes of critical structure
component Aluminium skin panel
Complex due to rivets and layers Approximately 2 mm to 5 mm thick
Load bearing components 10 to 20 mm thick Aluminium or steel Complex shape
Selected aspriority
Copyright TWI Ltd 2012
Aircraft Components
Images fromNDT Expert
Copyright TWI Ltd 2012
Aircraft Components Traditional maintenance scenario
Scheduled downtime Manual Inspection
Why guided waves? Volumetric defect detection
Few sensors low weight
Monitoring inaccessible components Remove need to dismantle
Copyright TWI Ltd 2012
Challenges Background Guided Wave Approach
Typically at low ultrasonic frequencies ~< 0.3MHz
Transducer array approach (heavy) Numerous transducers Numerous wires Done to reduce signal complexity
Required Development Medium frequency system (0.3 to 1MHz) Develop Transducers Computer Aided Defect Detection (Technique)
Copyright TWI Ltd 2012
Challenges Background Guided Wave Approach
Typically at low ultrasonic frequencies ~< 0.3MHz
Transducer array approach (heavy) Numerous transducers Numerous wires Done to reduce signal complexity
Required Development Medium frequency system (0.3 to 1MHz) Develop Transducers Computer Aided Defect Detection (Technique)
Very complex signals, poorfor human interpretation
Copyright TWI Ltd 2012
Challenges Background Guided Wave Approach
Typically at low ultrasonic frequencies ~< 0.3MHz
Transducer array approach (heavy) Numerous transducers Numerous wires Done to reduce signal complexity
Required Development Medium frequency system (0.3 to 1MHz) Develop Transducers Computer Aided Defect Detection (Technique)
Copyright TWI Ltd 2012
Development Summary Medium frequency system (0.3 to 1MHz)
Developing lab system Developing pure digital system
Develop Ultrasonics and Transducers Conducted vibrometry measurements Testing/developing a range of transducers
Neural Network Defect Detection (Technique) Assessed network types Chosen features Evaluated performance
Copyright TWI Ltd 2012
Develop Transducers Transducer development allows
Some improved signal clarity The use of medium frequency guided waves Reduced weight
Optimised transducers will lead to Less complex signals Greater defect sensitivity
Copyright TWI Ltd 2012
Thick Sample TrialsWith and without 5mm saw cut
Copyright TWI Ltd 2012
Thick Sample Trials
1)2)
3)4)
Copyright TWI Ltd 2012
Thick Sample Trials
Defect detection region of interest
Monolithic piezoceramic in-
plane shear transducer
10 mm thick aluminium
sample without defect
Spray deposited thin
matt white powder coating
Photograph of Structural Plate SampleView From Vibrometer Head Position
Photograph of Scanning Vibrometer
1-3 composite
compression transducer
Copyright TWI Ltd 2012
Thick Sample Trials
Point Disturbance In theHorizontal Axis
1-3 Composite Transducer
Side WallSurface Wave
Defect Detection Region Of Interest
Finite Element Analysis(In-plane Stress Magnitude)
Vibrometry Experiment(Out-of-plane Surface Velocity)
Copyright TWI Ltd 2012
Thick Sample Trials Fatigue crack growth up to 5mm
Pairs 1 & 2
Pair 4
Region For Potential Crack
Pair 3
Monolithic in-plane shear transducer
Receiver
Monolithic in-plane shear transducer
Transmitter
Monolithic in-plane shear transducer
Receiver
Monolithic in-plane shear transducer
Transmitter
Co-located 1-3 composite transmitter
and receivers
Copyright TWI Ltd 2012
Thick Sample Trials
Region For Potential Crack
With 1mm Notch
Pair 3 Pair 2 Pair 4Pair 1
Copyright TWI Ltd 2012
Thick Sample TrialsFatigue machine with three
point bending setupSample with crack initiation notch and
fixed transducer
Low frequency transmitter/receiver
and controlling laptop
High frequency control, arbitrary wave form
generator and received signal digitiser
High Frequency
Receiver Amplifier
High FrequencyTransmitter Amplifier
High FrequencyPower Supply
Board
Power Source
Copyright TWI Ltd 2012
Thick Sample Trials
A crack defect was slowly grown in a sample The defect size was monitored with manual
NDT Meanwhile, ultrasonic data was automatically
collected at regular intervals
Copyright TWI Ltd 2012
Thick Sample Trials
0
1
2
3
4
5
6
0 500 1000 1500 2000
Estim
ated
Sur
face
Cra
ck L
engt
h (m
m)
Fatigue, thousand cycles
Long Range Ultrasonic Data Collection Over Fatigue TestUltrasonic Testing
Copyright TWI Ltd 2012
Thick Sample Trials
0
1
2
3
4
5
6
0 500 1000 1500 2000
Estim
ated
Sur
face
Cra
ck L
engt
h (m
m)
Fatigue, thousand cycles
Long Range Ultrasonic Data Collection Over Fatigue TestUltrasonic Testing
No Defect SmallDefect
Significant Defect
Copyright TWI Ltd 2012
Thick Sample Trials
Notch
Die indicatingcrack
Copyright TWI Ltd 2012
Signal Bank Collected
0mm Crack 5mm Crack
Copyright TWI Ltd 2012
Technique Development Neural Network system
Pavlos Stavrou, CERETETH, Greece
What is an neural network ? An artificial neural network is an information
processing system whose structure andfunctionality is inspired by biological nervoussystems. Its key structural element is theneuron which is defined by its inputs, outputand activation function.
Copyright TWI Ltd 2012
Technique Development
How do neural networks work ? Neurons are structured in layers and information
propagates from the input to the output layer. Each input to a neuron is assigned a weight and along with the activation function they determine the information that is propagated to the output.
fw1
w2
wN
s1
s2
sN
y
y=f(wisi - )
Copyright TWI Ltd 2012
Technique Development What are the advantages of using neural
networks ? Distributed/Parallel information processing Robustness Training
How can NNs aid in defect detection ? Since LRU signals acquired from structures with
complex geometry are very complex, we need theprocessing and training capability of neuralnetworks to detect even the finest differences insignals in order to classify them accurately.
Copyright TWI Ltd 2012
Technique Development Steps followed for NN development
Each signal will be represented by its feature vector Feature Generation Feature Selection Neural Network Design Training the Neural Network Neural Network Validation and Error Probability Estimation Neural Network Refinement (Fine-Tuning) Field Testing and Evaluation
Features examined for aircraft component defect detection Estimated Central Frequency Central Frequency Deviation Bandwidth Dominant Pulse Power Standard Deviation Variance Covariance with reference non-defective