Intelligent Eddy Current Crack Detection System Design Based on Neuro-Fuzzy Logic Data fusion ECT signal processing Oct. 09 th, 2013 Baoguang Xu MASc. Concordia University Montreal 1
Outline Project description and goals Eddy current signal feature extraction and analysis using fuzzy logic. Signal de-noise Signal features and feature extraction Fuzzy logic Experiments and results Conclusions and Future work 2
1. Project description and goals 3
Team members 4
Project description and goals In aerospace industry especially in aircraft maintenance, ECT is performed manually. Time and cost consuming Result depends on human experience. Currently no signal recognition system is able to indicate crack features such as depth and shape etc. automatically In aircraft maintenance and manufacturing especially for quality control, ECT signal recognition largely relies on the properties of cracks The goal of the project--- to provide the aerospace industry with a user friendly, AI aided signal recognition system to obtain the cracks information automatically. 5
Program Design Diagram Known Signal input Signal de-noise Signal processing(feature extraction) Fuzzy logic training (ANFIS) Training Signal from unknown crack Signal de-noise Signal processing(feature extraction) Trained fuzzy logic Crack information output (depth, width shape etc.) 6
2. Eddy current signal feature extraction and analysis 7
Signal de-noise Filter choices High/Low pass filter Fourier transform. Wavelet transform. 8
Wavelet de-noise 1. Decompose signal into wavelet components (in which case noises are been separated). 2. Define the right wavelet C coefficient in order to miniature or remove noises. 3. Reconstruct processed signal by defining C coefficient [1] Filter S(signal ) Filter Filter ca1 Filter cd1 ca2 cd2 Filter Filter ca3 cd3 9
De-noise Result 10
Signal features and feature extraction 2 2 Z R X L Z : Impedance R: Resistance XL : Reactance Impedance of eddy current test Typical crack signal for differential probe 11
Signal features and feature extraction Before After 12
Approaches to obtain the crack information based on ECT signal Theoretical model Analytical Modeling Numerical Modeling Drawbacks: It is hard to establish the theoretical models due to the complexity of crack geometry and the non-accessible detailed coil information of the probe. Artificial intelligence Fuzzy logic Neural networks Drawbacks: Artificial intelligence techniques require large set of reliable data to train the system. 13
Fuzzy logic Fuzzy logic is known as an artificial intelligence tool to describe complicated physical phenomena and to anticipate the linear or nonlinear results based on collected input and output data [2]. Instead of using absolute 1 (true) and 0 (false) to make traditional logical decision, fuzzy logic introduces the concept of membership to describe how the input and output weight (in between 1 and 0) as a member in a certain membership. Members hip Function Fuzzy logic system Input Fuzzifier Fuzzy rules Duzzifier Output 14
ANFIS (Adaptive-Network-Based Fuzzy Inference System) Neuro-based fuzzy logic is inspired by neural network, similar to that of neural network which constitutes input and output mapping via their membership functions and related parameters [3]. Neuro mapping structure, Fuzzy rules samples after training 15
3. Experiments and Results 16
Hardware information Olympus: Nortec 500S Frequency Range 50 Hz -12 MHz Probe: differential reflection probe (PRL/500 khz - 3 MHz/D) Crack sample: Nortec TB-S1 Standard (deep notches: 8mil; 20mil; 40mil) Pictures come from:[5][6] 17
Hardware information National instruments: NI USB-6009 8 analog inputs (14-bit, 48 ks/s) 2 analog outputs (12-bit, 150 S/s); 12 digital I/O; 32-bit counter Compatible with LabVIEW, LabWindows /CVI, and Measurement Studio for Visual Studio.NET Pictures come from:[7] 18
Hardware Connection computer Data Acquisition Card ECT equipment 19
Testing Sample Material: Aluminum 7075-T6 notch crack sample 0.008 in. (0.203 mm), 0.020 in. (0.508 mm) and 0.040 in. (1.016 mm) 0.0315in. (0.80mm), 0.0591in. (1.50mm), 0.0787in. (2.0mm) 20
Experimental results 21
Experimental results 22
Experimental results 23
User interface ECT user friendly interface 24
Experimental results (angled crack) angled crack sample One important finding: The crack angle could have relation with the ratio of upper loop width and down side loop width, which will be investigated in the future 25
4. Conclusions and Future work 26
Conclusions The feature extraction algorism is able to process differential ECT signal Trained fuzzy logic possess an accurate result of crack depth predication with proper tuning method User interface is partial functional and is able to implement the feature extraction as well as the crack definition. 27
Future work More data collection Smaller cracks Cracks with complicated crack geometry Data from robotic scan system Theoretical model exploration Impedance trajectory simulation Data collection and training using theoretical modeling User interface improvement and implementation 28
References 1. Takagi T., Bowler J.R., Yoshida Y., Electromagnetic Nondestructive Evaluation, Volume 1, IOS Press, 1997.Smaller cracks 2. Wang, L. X. A Course in Fuzzy systems and Control, Prentice Hall PTR, 1997. 3. Jang, J.-S. R., ANFIS: Adaptive-Network-based Fuzzy Inference Systems, IEEE Transactions on Systems, Man, and Cybernetics, Vol. 23, No. 3, pp. 665-685, May 1993. 4. Garcia-Martin, J., Gomez-Gil, J. and Vazquez-Sanchez, E., "Non-Destructive Techniques Based on Eddy Current Testing", Sensors, vol.11, no.3, March 2011, pp. 2525-2565. 5. " Nortec 500Series Portable Eddy Current Flaw Detectors Operation Manual ", PN 7720140.00, Revision B, July 2013 6. " Olympus Eddy Current Probes catalog 7. National Instruments Product spec. http://sine.ni.com/nips/cds/view/p/lang/en/nid/201987 29
Thank you! Q&A 30