RADIOELECTRONIC AND COMPUTER SYSTEMS

The challenge in vibration analysis for helicopters is to automate anomaly detection to indicate possible equipment failure, especially in high-stakes scenarios where labelled failure data is rare and highly desirable for training modern models such as convolutional neural networks or autoencoders which require a lot of such data and are often used for this kind of analysis. The study focuses on vibration data generated by helicopters, focusing on anomalies in these vibrations that may indicate imminent equipment failure. These vibrations are often complex and change under different operating conditions, making them a valuable source of diagnostic information. Objective and Approach. This study presents a novel model for detecting abnormal vibrations in helicopter operations, minimising dependency on failure data. The model uses Fast Fourier Transform (FFT) and convolutional techniques to reduce high-dimensional vibration data into a three-dimensional vector, enabling effective anomaly detection through distribution metrics and Z-score-based thresholding. Case Study. The model was evaluated using a real-world dataset of helicopter vibration signals. The study successfully differentiated between normal and abnormal conditions without relying on explicit failure labels, validating its applicability in maintenance scenarios where early fault detection is critical. Method. Model was developed using Fast Fourier Transform (FFT) and convolutions to reduce vibration data to a low-dimensional vector representation that allows anomaly detection using distribution metrics. This method increases adaptability and reduces data requirements, making it suitable for expensive equipment with limited or no fault data. In addition, this model can be very flexible and extended to a model with additional training. It uses the Fast Fourier Transform as the first step to get the frequency components of a signal buried in vibrations. After FFT, we can represent the vibration data as a multidimensional vector. An example can consist of around 45000 dimensions after that vector simplifies to reduce the dimensions. The model uses convolutions to get only a 3-dimensional vector from the 45,000-dimensional vector. After convolutions, the data saves a clear correlation, and it is still possible to represent the vibration sample as a vector in 3-dimensional space and save its properties. Three dimensions were chosen for visualisation comfort, and it still works correctly. These vectors are located in space in two separate clusters for 0 and 1 corresponding labels, and they can be used for the next iteration of analysis. On the next iteration, the model gets a mean from x,y, and z and checks the distributions for class 0 and class 1. This step is possible because values after all convolutions for x,y, and z are close to each other. As distributions of these vectors are close to a bell shape, the model uses distribution metrics to predict fault by using a Z-score of the distribution of normal vectors. In this case, the model is close to the zero-shot learning concept because the model requires only data about expected behaviour. Results. The developed model showed high accuracy at a particular configuration when using different matrix convolutions and a different approach to setting the limit of anomalous values. At the final stage of the testing, the model reached an accuracy of 99.568%, which, under specific conditions, is not considered an overfitted model since the initial input parameters were clearly defined and limited by specific data. The model also demonstrated that when changing the input conditions, the model can be easily configured to new parameters. Discussion. The proposed model proved highly adaptable and accurate in detecting anomalies with minimal historical data. While it showed robustness across different configurations, future improvements could focus on refining sensitivity thresholds and integrating additional predictive techniques for enhanced performance. Conclusions. Tests validate the model's adaptability and scalability, recommending its use for fault detection in inexpensive, data-limited environments like helicopters.

vibration analysis; helicopter; anomaly detection; convolutional model; fast fourier transform; zero-shot learning

Li, W., Huang, R., Li, J., Liao, Y., Chen, Z., He, G., Yan, R., & Gryllias, K. A perspective survey on deep transfer learning for fault diagnosis in industrial scenarios: theories, applications and challenges. Mech. Syst. Signal Process, 2022, vol. 167, part. A, article no. 108487. DOI: 10.1016/j.ymssp.2021.108487.

Zhao, Y. P., & Bin Chen, Y. Extreme learning machine based transfer learning for aero engine fault diagnosis. Aerosp. Sci. Technol., 2022, vol. 121, article no. 107311. DOI: 10.1016/J.AST.2021.107311.

Huang, P., Chen, X., Chai, Y., Ma, L. A unified framework of fault detection and diagnosis based on fractional-order chaos system. Aerosp. Sci. Technol., 2022, vol. 130, article no. 107871, DOI: 10.1016/J.AST.2022.107871.

Han, L., Chen, C., Guo, T., Lu, C., Fei, C., Zhao, Y., & Hu, Y. Probability-based service safety life prediction approach of raw and treated turbine blades regarding combined cycle fatigue. Aerosp. Sci. Technol., 2021, vol. 110, article no. 106513. DOI: 10.1016/J.AST.2021.106513.

Li, X., Jiang, H., Liu, Y., Wang, T., & Li, Z. An integrated deep multiscale feature fusion network for aeroengine remaining useful life prediction with multisensor data. Knowledge-Based Syst., 2022, vol. 235. DOI: 10.1016/j.knosys.2021.107652.

Lu, C., Li, H., Han, L., Keshtegar, B., & Fei, C. W. Bi-iterative moving enhanced model for probability-based transient LCF life prediction of turbine blisk. Aerosp. Sci. Technol., 2023, vol. 132, article no. 107998, DOI: 10.1016/J.AST.2022.107998.

Wu, J., Hu, C., Sun, C., Zhao, Z., Yan, R., & Chen, X. Helicopter transmission system anomaly detection in variable flight regimes with decoupling variational autoencoder. Aerospace Science and Technology, 2024, vol. 144, article no. 108764. DOI: 10.1016/j.ast.2023.108764.

Li, W., Huang, R., Li, J., Liao, Y., Chen, Z., He, G., Yan, R., & Gryllias, K. A perspective survey on deep transfer learning for fault diagnosis in industrial scenarios: Theories, applications and challenges. Mechanical Systems and Signal Processing, 2022, vol. 167, Part A, article no. 108487. DOI: 10.1016/j.ymssp.2021.108487.

Chu, T., Nguyen, T., Yoo, H., & Wang, J. A review of vibration analysis and its applications. Heliyon, 2024, vol. 10, iss. 5, article no. e26282. DOI: 10.1016/j.heliyon.2024.e26282.

Avci, O., Abdeljaber, O., Kiranyaz, S., Hussein, M., Gabbouj, M. & Inman, D. J. A review of vibration-based damage detection in civil structures: From traditional methods to Machine Learning and Deep Learning applications. Mechanical Systems and Signal Processing, 2021, vol. 147, article no. 107077. DOI: 10.1016/j.ymssp.2020.107077.

Lv, Y., Zhao, W., Zhao, Z., Li, W., & Ng, K. K. H. Vibration signal-based early fault prognosis: status quo and applications. Adv. Eng. Inf., 2022, vol. 52, article no. 101609. DOI: 10.1016/j.aei.2022.101609.

Khan, M. A., Akhtar, K., Ahmad, N., Shah, F., & Khattak, N. Vibration analysis of damaged and undamaged steel structure systems: cantilever column and frame. Earthq. Eng. Eng. Vib., 2020, vol. 19, pp. 725-737. DOI: 10.1007/s11803-020-0591-9.

Gholamy, A., Kreinovich, V., & Kosheleva, O. Why 70/30 or 80/20 Relation between Training and Testing Sets: A Pedagogical Explanation. Departmental Technical Reports (CS), 2018. Available at: https://scholarworks.utep.edu/cs_techrep/1209. (accessed 10.10.2024).

Lim, W., Yong, K. S. C., Lau, B. T., & Tan, C. C. L. Future of generative adversarial networks (GAN) for anomaly detection in network security: A review. Computers & Security, 2024, vol. 139, article no. 103733. DOI: 10.1016/j.cose.2024.103733.

Vladov, S., Ścisło, Ł., Szczepanik-Ścisło, N., Sachenko, A., Perzyński, T., Vasylenko, V., & Vysotska, V. Helicopter turboshaft engines’ neural network system for monitoring sensor failures. Sensors, 2025, vol. 25, article no. 990. DOI: 10.3390/s25040990.

Vladov, S., Vysotska, V., Sokurenko, V., Muzychuk, O., Nazarkevych, M., & Lytvyn, V. Neural network system for predicting anomalous data in applied sensor systems. Appl. Syst. Innov., 2024, vol. 7, article no. 88. DOI: 10.3390/asi7050088.

Ballarin, P., Sala, G., Macchi, M., Roda, I., Baldi, A., & Airoldi, A. Application of artificial neural networks to a model of a helicopter rotor blade for damage identification in realistic load conditions. Sensors, 2024, vol. 24, article no. 5411. DOI: 10.3390/s24165411.

Sáenz-Hernández, C., Cuadros, R., Rodríguez, J., Rativa, E., Linares-Vásquez, M., Donoso, Y., & Lozano, C. Development of a test bench for fault diagnosis in the caution and warning panels of the UH-60 helicopter. Eng, 2025, vol. 6, article no. 101. DOI: 10.3390/eng6050101.

Roshchupkin, O., & Pavlenko, I. Modern methods for diagnosing faults in rotor systems: A comprehensive review and prospects for AI-based expert systems. Appl. Sci., 2025, vol. 15, article no. 5998. DOI: 10.3390/app15115998.

Hodlevskyi, Y., Vakaliuk, T. Vibration dataset. Zenodo, 2026. DOI: 10.5281/zenodo.19475281.