Carbon/Epoxy Laminate Compression After Impact Load Carbon/Epoxy Laminate Compression After Impact Load Prediction from Ultrasonic C-Scan DataPrediction from Ultrasonic C-Scan Data

Eric v. K. Hill, Christopher D. Hess and Yi ZhaoEric v. K. Hill, Christopher D. Hess and Yi Zhao

OBJECTIVES

•Three sets of 3.5 x 6 inch 16-ply AS4/3501-5A carbon/epoxy coupons impacted from 0-20 ft-lbf with 5/8 inch diameter hemispherical tup to create barely visible impact damage (BVID)

•Back-propagation neural network (BPNN) prediction of compression after impact (CAI) load from transformed ultrasonic (UT) C-scan image

•Goal:Goal: Worst case prediction error within ±15%±15%

APPROACH/TECHNICAL CHALLENGES

• AE data too noisy: Train BPNN using 50 data points representing column summation data from UT C-scan image and known CAI loads as input

• Test BPNN using column summation UT C-scan image to predict CAI loads on remaining coupons

ACCOMPLISHMENTS/RESULTS• UT image data alone used to predict ultimate

compressive strengths with worst case errors worst case errors of -12.12%, 16.62%, and -11.83% for the three sets-12.12%, 16.62%, and -11.83% for the three sets• BPNN able to predict accurately without known predict accurately without known

impact energyimpact energy – valid for real world applications such as impact damaged aircraft wings

C/Ep Coupon C/Ep Coupon in Boeing in Boeing BS-7260 BS-7260

Compression Compression After Impact After Impact Test Fixture Test Fixture with Three with Three Acoustic Acoustic Emission Emission

Transducers Transducers AttachedAttached

Instron Dynatup Instron Dynatup 9250 Calibrated 9250 Calibrated

ImpactorImpactor

Delaminations in Delaminations in Coupon Due to Coupon Due to Impact DamageImpact Damage

MATLAB Data TransformationMATLAB Data Transformation

• Pixel color and location is represented by a matrix array of numbers (0-16)

• Numerical values represent hue color• Image data summed and normalized in the

column direction • 50-100 data points surrounding the

maximum used as inputs to BPNNUltraPAC II C-Scan Imaging System:• Water Couplant Immersion• 5 MHz Unfocused Transducer

16 Color Format0-15 Color Format0-15 Color Format Digital Representation of 0-15 Color FormatDigital Representation of 0-15 Color Format

 Data Set  Specimen Impact Energy (ft-lbf)Compressive

Load (lbf)Predicted Compressive

Load (lbf)% Error

Training

A2 0 2865.6 2865.60 0.00

A3 2.23 6531.9 6531.90 0.00

A5 21.43 3910.1 3910.10 0.00

A4 20.2 3042.4 3042.40 0.00

TestingA6 20.75 4174.8 4492.73 7.62

A1 0 4936.5 4338.07 -12.12-12.12

BPNN Predictions for “Batch A” CouponsBPNN Predictions for “Batch A” Coupons

Optimized BPNN SettingsOptimized BPNN Settings

Digital Ultrasonic Digital Ultrasonic C-ScanC-Scan

Image DataImage Data

PredictedPredictedCAI LoadCAI Load

NeuralWorksNeuralWorksProfessional II/PLUSProfessional II/PLUS® ®

SoftwareSoftware

Summary of BPNN Training and Test Summary of BPNN Training and Test ResultsResults

Worst Case ErrorWorst Case Error