A Combined Finite Element Analysis and Artificial Neural Network Approach for Diagnostics of Building Cross-Sections Weakened with Stress Concentrators

Introduction. The study is dedicated to developing a new method for identifying defects in building structures with stress concentrators. The method is based on the integration of shadow ultrasonic testing with deep learning algorithms, which would enable accurate diagnosis with reliably identifying the geometric characteristics of defects.

Materials and Methods. A finite element model of an area with an angular point and damping layers made of metal with a flexible coating was used. An ultrasonic actuator and receiver were placed on the opposite edges. Numerical experiments with changes in the geometry and materials of the area and defect parameters were conducted on a distributed computing system. The resulting signals were converted into spectrograms which were used in order to train a convolutional neural network that establishes a connection between the spectrogram and the defect parameters.

Results. An extensive dataset of spectrograms has been formed. The trained neural network has displayed the ability to accurately identify the key defect parameters based on a spectrogram such as size, position, and orientation. Verification of the method has shown that it outperforms the traditional methods of ultrasonic signal analysis in terms of its accuracy and speed.

Discussion and Conclusion. The hybrid approach for non-destructive testing in complex geometric conditions has been proven to be effective. The major advantage is automated and intelligent data analysis reducing a degree of subjectivity. The practical significance is the creation of a prototype adaptive diagnostic system. Prospects are related to further training on experimental data and integration into portable systems for monitoring structures. 

1. Alexandrov VM, Smetanin BI, Sobol BV Thin Stress Concentrators in Elastic Bodies. Moscow: Nauka, 1993. 224 p. (In Russ.)

2. Sobol BV, Rashidova EV, Ivashchenko VV Equilibrium State of a Straight-Line Internal Crack Near the Corner Point of an Elastic Region Reinforced along the Contour. Bulletin of Perm National Research Polytechnic University. Mechanics. 2025; 2:100–110. (In Russ.) https://doi.org/10.15593/perm.mech/2025.2.09

3. Vatulyan AO, Belyak OA, Sukhov DYu, Yavruyan OV Inverse and Ill-Posed Problems: Textbook. Rostov-on-Don: Publishing House of the Southern Federal University; 2011. 232 p. (In Russ.)

4. Liu S., Wang Y., Yang X., Lei B., Liu L., Xiang L.S. et al. Deep Learning in Medical Ultrasound Analysis: A Review. Engineering. 2019;5(2):261–275. https://doi.org/10.1016/j.eng.2018.11.020

5. Van Sloun R.J.G., Cohen R., Eldar Y.C. Deep Learning in Ultrasound Imaging. Proceedings of the IEEE. 2020;108(1):11–29. https://doi.org/10.1109/JPROC.2019.2932116

6. Aleshin NP, Krysko NV, Kusyy AG, Skrynnikov SV Studying the Detectability of Place Surface Defects by Ultrasonic Method Using Rayleigh Waves. Defectoscopy. 2021; 5:22-30. 2021;6:26–34. (In Russ.) https://doi.org/10.31857/S0130308221060038

7. Dung C.V., Anh L.D. Autonomous concrete crack detection using deep fully convolutional neural network. Automation in Construction. 2019;99(3):52–58. https://doi.org/10.1016/j.autcon.2018.11.028

8. Ioffe S., Szegedy C. Batch normalization: Accelerating deep network training by reducing internal covariate shift. ArXiv. Computer Science. 2015. https://doi.org/10.48550/arXiv.1502.03167

9. Turco E. Tools for the numerical solution of inverse problems in structural mechanics: review and research perspectives. European Journal of Environmental and Civil Engineering. 2017;21(5):509–554. https://doi.org/10.1080/19648189.2015.1134673

10. Vasiliev PV, Senichev AV Application of Neural Network Technologies in the Problem of Surface Defect Inspection. Bulletin of Higher Educational Institutions. North Caucasus region. Technical Sciences. 2020;1:33–40. (In Russ.) http://dx.doi.org/10.17213/1560-3644-2020-1-33-40

11. Soloviev A., Sobol B., Vasiliev P. Identification of Defects in Pavement Images Using Deep Convolutional Neural Networks. In: Advanced Materials. Cham: Springer; 2019. Pp. 615–626. https://doi.org/10.1007/978-3-030-19894-7_46

12. Jiang L.L., Maskell D.L. Automatic fault detection and diagnosis for photovoltaic systems using combined artificial neural network and analytical based methods. In: International Joint Conference on Neural Networks (IJCNN). Killarney: IEEE; 2015. https://doi.org/10.1109/IJCNN.2015.7280498