A Recurrent Neural Network Classifier for Ultrasonic NDE Applications
- Resource Type
- Conference
- Authors
- Marino, Michael; Virupakshappa, Kushal; Oruklu, Erdal
- Source
- 2018 IEEE International Ultrasonics Symposium (IUS) Ultrasonics Symposium (IUS), 2018 IEEE International. :1-4 Oct, 2018
- Subject
- Aerospace
Bioengineering
Communication, Networking and Broadcast Technologies
Components, Circuits, Devices and Systems
Fields, Waves and Electromagnetics
Signal Processing and Analysis
Band-pass filters
Detectors
RNN
BRNN
LSTM
Wavelet Transform
NDE
flaw detection
- Language
- ISSN
- 1948-5727
This work presents a classifier architecture for N on-Destructive Evaluation (NDE)applications which can robustly detect the presence and location of flaws using Wavelet Transforms (WT)and Recurrent Neural Networks (RNN). A pre-processing WT decomposition is used as a feature extractor prior to the RNN. WT can analyze both time and frequency information simultaneously and can filter out higher frequency clutter via the selection of the appropriate wavelet and number of decomposition levels. Two popular RNN classifiers were investigated. Bidirectional Recurrent Neural Networks (BRNN)and Long Short Term Memory (LSTM). Simulation results confirm that the proposed architecture offers highly reliable flaw detection and localization with significant Flaw to Clutter Ratio (FCR)enhancements. This architecture also performs well when multiple adj acent flaws exist.