Use of Artificial Intelligence to Filter Optical Images of Wheel Surface Defects and Reconstruct Missing Data Based on Standardized Wheel Rim Geometry

Introduction. Modern trends in the development of railway transport, associated with increasing speeds and loads, impose stricter requirements on technical diagnostic systems. Continuous monitoring of wheelset condition is of critical importance, since their defects directly affect operational safety. Conventional diagnostic methods require trains to be taken out of service for inspection, resulting in limited efficiency. Existing optical inspection systems often fail to operate reliably in conditions of high-speed traffic and intense electromagnetic interference, significantly impeding the possibility of continuous monitoring.
Aim. To develop and test an automated optical system for real-time high-speed monitoring of rolling surface geometry without taking the rolling stock out of service.
Materials and methods. The proposed system is installed beneath the rail head and includes an array of eight thermally stabilized pulsed lasers synchronized with high-sensitivity CMOS arrays. The system is based on a dark-field imaging scheme that selectively captures signals from defects against the background of interference. To minimize interference-related distortions, an original algorithm for sequential separate scanning of wheel elements is used. The data is processed using AI-based algorithms. The system was experimentally validated both in laboratory conditions on a test bench and during field tests on a railway track.
Results. The experiments confirmed the stable operation of the system at speeds up to 100 km/h, as well as its high noise immunity to electromagnetic interference. AI-based data processing algorithms enable the determination of geometric parameters of the wheel profile with an accuracy of 0.2 mm. The use of artificial intelligence makes it possible to effectively suppress residual interference detected by the matrix.
Conclusion. The developed system offers significant advantages over conventional diagnostic methods by enabling real-time inspection without interrupting railway operations. Unlike existing analogues, the proposed system is adapted to operation in difficult conditions of high speeds and strong interference. Its implementation forms a basis for the transition to predictive analytics and maintenance strategies based on the actual condition of the railway industry.

1. Wang N., Jia L., Zhang H., Qin Y., Wang Z., Zhao X. Manifold-Contrastive Broad Learning System for Wheelset Bearing Fault Diagnosis. IEEE Trans. on Intelligent Transportation Systems. 2023, vol. 24, no. 9, pp. 9886–9900. doi: 10.1109/TITS.2023.3274256

2. Arinushkina K. G., Adadurov A. S., Savelev I. Yu., Davydov V. V. Features of the Operation of a Laser Profilometer in an Automated Rolling Stock Control System. St Petersburg State Polytechnical University J. Physics and Mathematics. 2023, vol. 16, no. S3.2, pp. 56–62. DOI: 10.18721/JPM.163.209

3. Boonmeeprakob K., Phanpumnak A., Punwong C., Chana S., Thavarungkul P., Kanatharana P., Buranachai C. Portable Contactless Caliper. Appl. Optics. 2024, vol. 63, no. 1, pp. 122–128. doi: 10.1364/AO.507349

4. Polyakov V., Nearing M. A Simple Automated Laser Profile Meter. Soil Science Society of America J. 2019, vol. 83, no. 2, pp. 327–331. doi: 10.2136/sssaj2018.10.0378

5. Yan Z., Zhu H., Zhang H., Wan H., Liu B. Abnormal Noise Monitoring of Subway Vehicles Based on Combined Acoustic Features. Applied Acoustics. 2022, vol. 197, art. no. 108951. doi: 10.1016/j.apacoust.2022.108951

6. Jelila Y. D., Pamuła W. Detection of Tram Wheel Faults Using MEMS-Based Sensors. Sensors. 2022, vol. 22, no. 17, art. no. 6373. doi: 10.3390/s22176373

7. Emoto T., Ravankar A. A., Ravankar A., Emaru T., Kobayashi Y. Wheel Surface Defect Detection Method Using Laser Sensor and Machine Vision. Proc. of 62nd Annual Conf. of the Society of Instrument and Control Engineers, , Tsu, Japan, 06–09 Sept. 2023. IEEE, 2023, vol. 2023, pp. 1194–1199. doi: 10.23919/SICE59929.2023.10354134

8. Arinushkina K. G. Overview of Fiber Optic Communications in Railway Transport: Challenges and Impact of Processing Time. Proc. of the 2024 Conf. of Young Researchers in Electrical and Electronic Engineering, St Petersburg, 29–31 Jan. 2024. IEEE, 2024, pp. 870–874. doi: 10.1109/ElCon61730.2024.10468372

9. Zheng F., Zhang B., Gao R., Feng Q. A High-Precision Method for Dynamically Measuring Train Wheel Diameter Using Three Laser Displacement Transducers. Sensors. 2019, vol. 19, no. 19, art. no. 4148. doi: 10.3390/s19194148

10. Dubnishchev Yu. N., Belousov P. Ya., Sotnikov V. V., Belousova O. P. Optical Control of the Radius of a Wheel Rolling on a Rail. Optoelectronics, Instrumentation and Data Processing. 2012, vol. 48, no. 1, pp. 75‒80. doi: 10.3103/S8756699012010116

11. Boronenko Yu. P., Tret’yakov A. V., Rakhimov R. V., Zimakova M. V., Nekrasova A. V., Tret’yakov O. А. Monitoring the Technical Condition of the Railway Track Using the Method of Continuous Recording of Dynamic Processes Occurring Due to The Interaction Between Rolling Stock and Railway Track. Bulletin of Scientific Research Results. 2021, vol. 3, pp. 66–82. (In Russ.) doi: 10.20295/2223-9987-2021-3-66-82

12. Zheng Z., Song D., Xu X., Lei L. A Fault Diagnosis Method of Bogie Axle Box Bearing Based on Spectrum Whitening Demodulation. Sensors. 2020, vol. 20, no. 24, art. no. 7155. doi: 10.3390/s20247155

13. Yan B., Xiao R., Cai Z., Sun H., Chen Q. Digitally Controll ed Frequency-Locked Circuit for Ultrastable Laser. Chinese J. of Lasers. 2024, vol. 51, no. 13. Art. no. 130101. doi: 10.3788/CJL240450

14. Faghih-Roohi S., Hajizadeh S., Núñez A., Babuska R., De Schutter B. Deep Convolutional Neural Net-works for Detection of Rail Surface Defects. Intern. Joint Conf. on Neural Networks, Vancouver, Canada, 24–29 July 2016. IEEE, 2016, pp. 2584–2589. doi: 10.1109/IJCNN.2016.7727522

15. Bešinović N., De Donato L., Flammini F., Goverde R. M. P., Lin Zh., Liu R., Marrone S., Nardone R., Tang T., Vittorini V. Artificial Intelligence in Railway Transport: Taxonomy, Regulations and Applications. IEEE Trans. on Intelligent Transportation Systems. 2021, vol. 23, no. 9, pp. 14011–14024. doi: 10.1109/TITS.2021.3131637

16. Adadurov A. S., Savelev I. Yu., Chepulis M. A., Arinushkina K. G. Development of an Advanced Algorithm for Calculating and Assessing the Wheel Pair Geometry in the CONLOC system. Russian Railway Science J. 2023, vol. 82, iss. 3, pp. 247–256. (In Russ.) doi: 10.21780/2223-9731-2023-82-3-247-256

17. Adadurov A. S., Savelev I. Y., Tsomaev P. I., Arinushkina K. G. Automated Non-contact Measurement of Geometric Parameters of Wheels for Safe Movement of Rolling Stock. Proc. of the Ninth Intern. Sci. Conf. Intelligent Information Technologies for Industry. Springer, Cham, 2026, vol. 1, pp. 65‒76. doi: 10.1007/978-3-032-13615-2_7