502

TransWind: A Vision Transformer Framework for Wind Turbine Fault Diagnosis on Supervisory Control and Data Acquisition System

Abdullah Saeed Alwadie(1), Sana Yasin(2), Muhammad Irfan(1), Umar Draz(3), Tariq Ali(4)

1. Electrical Engineering Department, College of Engineering, Najran University, Najran 61441; Saudi Arabia.
2. Department of Computer Science, Faculty of Computing, University of Okara, Pakistan.
3. Department of Computer Science, Faculty of Computing and Information Technology, University of Sahiwal, Pakistan.
4. Artificial Intelligence and Sensing Technologies (AIST) Research Center, University of Tabuk, Saudi Arabia.

2026-02-15

Accurate fault detection in wind turbines is essentialfor maximizing operational efficiency and reducing maintenance expenditures. This paper presents TransWind, a novel Vision Transformer (ViT)-based framework designed specifically for analyzing SCADA data to pinpoint and diagnose issues in wind turbines. Unlike traditional machine learning models, TransWind leverages the attention mechanisms of ViTs to capture complex temporal and spatial relationships within SCADA time-series data, including parameters for example rotational speed, generator temperature, and electricity output. This unique capability allows for precise identification of anomalies and their underlying causes. The innovation of TransWind lies in its capacity to integrate explainable AI (XAI) techniques, including Self-Attention Attribution, to provide transparency in fault predictions, enabling maintenance teams to focus on critical system elements. The proposed framework will assessed on publicly available SCADA datasets, demonstrating a fault detection accuracy improvement of 8-12% compared to state-of-the-art models. Furthermore, TransWind exhibits robustness in handling noisy and incomplete SCADA data, a common challenge in real-world deployments. This research highlights the transformative potential of transformer-based architectures in renewable energy fault diagnostics. By enhancing detection accuracy and interpretability, TransWind offers a scalable solution for predictive maintenance, reducing turbine downtime and operational costs while advancing AI-driven sustainability in wind energy systems.

Key words: Vision Transformers (ViTs), SCADA Data Analysis, Wind Turbine Fault Diagnosis, Explainable Artificial Intelligence, Predictive Maintenance.

[1] Khan, P. W., Yeun, C. Y., & Byun, Y. C. (2023). Faultdetection of wind turbines using SCADA data and geneticalgorithm-based ensemble learning. Engineering FailureAnalysis, 148, 107209.

[2] Wu, Z., Li, Y., & Wang, P. (2024). A hierarchical modelingstrategy for condition monitoring and fault diagnosis of windturbine using SCADA data. Measurement, 227, 114325.

[3] Dao, P. B., Barszcz, T., & Staszewski, W. J. (2024). Anomalydetection of wind turbines based on stationarity analysis ofSCADA data. Renewable Energy, 232, 121076.

[4] Maldonado-Correa, J., Torres-Cabrera, J., Martín-Martínez, S.,Artigao, E., & Gómez-Lázaro, E. (2024). Wind turbine faultdetection based on the transformer model using SCADAdata. Engineering Failure Analysis, 162, 108354.

[5] Wang, M. H., Lu, S. D., Hsieh, C. C., & Hung, C. C. (2022).Fault detection of wind turbine blades using multi-channelCNN. Sustainability, 14(3), 1781.

[6] Pandit, R., & Wang, J. (2024). A comprehensive review onenhancing wind turbine applications with advanced SCADAdata analytics and practical insights. IET Renewable PowerGeneration, 18(4), 722-742.

[7] Maldonado-Correa, J., Valdiviezo-Condolo, M., Artigao, E.,Martín-Martínez, S., & Gómez-Lázaro, E. (2024). Classificationof Highly Imbalanced Supervisory Control and DataAcquisition Data for Fault Detection of Wind TurbineGenerators. Energies, 17(7), 1590.

[8] Vásquez-Rodríguez, G., & Maldonado-Correa, J. (2024, July).Anomaly-based fault detection in wind turbines usingunsupervised learning: a comparative study. In IOP ConferenceSeries: Earth and Environmental Science (Vol. 1370, No. 1, p.012005). IOP Publishing.

[9] Khan, P. W., Yeun, C. Y., & Byun, Y. C. (2023). Fault detectionof wind turbines using SCADA data and genetic algorithm-based ensemble learning. Engineering Failure Analysis, 148,107209.

[10] Castellani, F., Natili, F., Astolfi, D., & Vidal, Y. (2024). Windturbine gearbox condition monitoring through the sequentialanalysis of industrial SCADA and vibration data. EnergyReports, 12, 750-761.

[11] Qiao, L., Zhang, Y., & Wang, Q. (2023). Fault detection in windturbine generators using a meta-learning-based convolutionalneural network. Mechanical Systems and SignalProcessing, 200, 110528.

[12] Wang, T., & Yin, L. (2024). A hybrid 3DSE-CNN-2DLSTMmodel for compound fault detection of wind turbines. ExpertSystems with Applications, 242, 122776.

[13] de Lima Munguba, C. F., Ochoa, A. A. V., Leite, G. D. N. P., daCosta, A. C. A., da Costa, J. Â. P., de Menezes, F. D., ... & deSouza, M. G. G. (2024). Fault detection framework in windturbine pitch systems using machine learning: Development,validation, and results. Engineering Applications of ArtificialIntelligence, 138, 109307.

[14] Milani, A. E., Zappalá, D., Castellani, F., & Watson, S. (2024,June). Boosting field data using synthetic SCADA datasets forwind turbine condition monitoring. In Journal of Physics:Conference Series (Vol. 2767, No. 3, p. 032033). IOPPublishing.

[15] Wang, M. H., Lu, S. D., Hsieh, C. C., & Hung, C. C. (2022).Fault detection of wind turbine blades using multi-channelCNN. Sustainability, 14(3), 1781.

[16] Irfan, M., Yasin, S., Draz, U., Ali, T., Yasin, I., Kareri, T., &Rahman, S. (2025). Revolutionizing wind turbine faultdiagnosis on supervisory control and data acquisition systemwith transparent artificial intelligence. International Journal ofGreen Energy, 1-17.

[17] Zhang, K., Tang, B., Deng, L., & Yu, X. (2021). Fault detectionof wind turbines by subspace reconstruction-based robust kernelprincipal component analysis. IEEE Transactions onInstrumentation and Measurement, 70, 1-11.

[18] He, Y., Liu, J., Wu, S., & Wang, X. (2023). Conditionmonitoring and fault detection of wind turbine driveline with theimplementation of deep residual long short-term memorynetwork. IEEE Sensors Journal, 23(12), 13360-13376.