Signal processing at eddy current defectoscopy of composites using artificial neural networks

1Antoniuk, IN, 1Antoniuk, OP
1Oles Honchar Dnipro National University, Dnipro, Ukraine
Kosm. nauka tehnol. 2002, 8 ;(Supplement1):101-105
Publication Language: Russian
Abstract: 
Carbon containing composite materials are widely implemented in different constructions of airspace technique due to their unique physico-mechanical properties. Last few years there has been intensive development and implementation of artificial neural networks for processing of signals, recognition and correction of images. This paper is dedicated to the creation of the neural network and to the elaboration of the algorithms of network's teaching for recognition of signals from defects and drawbacks that appear during eddy current testing of carbon fiber composites. Composite materials on the basis of carbon tissue have considerable roughness of the surface. That is why while scanning the surface of the material with an eddy current transformer, which has field centered in small volume, there often occur casual inclinations of the transformer. This makes false impulses which are comparable in amplitude and width with modulation impulses of the defect and they are frequently even bigger than impulses of surface cracks. Changes of the gap between the eddy current transformer and a surface of a testing material, casual inclinations of the transformer during scanning forms drawback impulses. These are the most essential preventing factors. The obtained results let to make correction of these factors and considerably improve reliance of defectoscopy.
References: 
1. Kawada Akira, Hayash Sigeyu. Methods for Diagnostic of Works Using Computer-Simulated Neural Networks. Gien, No. 80, 16—20 (1994).
2. Roy Amitava, Barat P., De Swapan Kumar. Material classification throughout neural network. Ultrasonics, 33, 175—180 (1995).
3. Wang Suju. Use of Neural Networks in Nondestructive Tests with the Help of Eddy Currents. Narjing HongKong Daxue Huebao (J. Narjing Univ. Aerocom. and Astronaut.), 27 (5), 696—700 (1995).
4. Lei Yinishao, Ma Xinshen. Use of Intelligent Neural Networks in Eddy-Current Tests. Wasan Jinance (Nondestructive Testing), 16 (2), 31—33, 57 (1994).
5. Kreis Thomas, Juptner Werner, Biedermann Ralf. Neural Network Approach to Holographic Nondestructive Testing. Appl. Optics, 34 (8), 1407—1415 (1995).
6. Koh Chan Scop, Mohammed Osana A., Hahn Song-yop. Detection of Magnetic Materials Using Artificial Neural Network with Modified Simulated Annealing. IEEE Trans. Magn., 30 (5), pt. 2, 3644—3647 (1994).
7. Wetzlar Dietmar. Neuronale Netze in der Megtechnic. Techn. Mess., 62 (3), 87—90 (1995) [in German].
8. Cai Yu-dong, Vao Lin-Chang. Utilization of Artificial Neural Network in Calibration of Transducers. Chin. I.: Sci. Instrum., 15 (3), 299—302 (1994).
9. Pham D. T., Bayro-Corrochano E. I. Neural Classifiers for Automated Inspection. Inst. Mech. Eng., 208 (2), 83—89 (1994).
10. Enokizino M., Todaka T., Akita M., Nagata S. Rotational Magnetic Flux Sensor with Neural Network for Nondestructuve Testing. IEEE Trans. Magn., 29 (6), pt. 1, 3195—3197 (1993).
11. Pristavka A. P., Pristavka P. A., Smirnov S. A. Statistical analysis in the ASTD: Time series, 69—75 (RIO DSU, Dnipropetrovsk, 2000) [in Ukrainian].

12. Khandetskii V. S., Antonyuk I. N. In: Defektoskopija, No. 4 (2001) [in Russian].