A method for identifying structural health anomalies in wind power hybrid towers without labels

By combining unsupervised deep learning and adaptive thresholding mechanisms with working condition-assisted feature fusion, the problem of early damage identification in wind power hybrid tower structures is solved, achieving efficient health monitoring and reduced false alarm rate, and adapting to the anomaly identification of wind power hybrid tower structures under different working conditions.

CN122087629APending Publication Date: 2026-05-26国能康平发电有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
国能康平发电有限公司
Filing Date
2025-12-19
Publication Date
2026-05-26

AI Technical Summary

Technical Problem

Existing technologies struggle to effectively identify early damage to wind turbine hybrid tower structures in the absence of fault labels, and are prone to false alarms under high wind speeds or heavy loads, making reliable health monitoring of concrete-steel hybrid tower structures impossible.

Method used

An unsupervised deep learning method is adopted, combined with working condition-assisted feature fusion and adaptive threshold mechanism. Through multi-source monitoring data acquisition, data preprocessing and deep spatiotemporal autoencoder model, anomalies in mixed tower structures are identified, the false alarm rate is reduced and the system adapts to changes in working conditions.

Benefits of technology

It enables early damage identification without fault tags, reduces false alarm rate, improves the reliability of health monitoring and operation optimization capabilities of wind power hybrid tower structures, and adapts to deployment in different wind farms and seasons.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122087629A_ABST
    Figure CN122087629A_ABST
Patent Text Reader

Abstract

This invention discloses a label-free method for identifying structural health anomalies in wind turbine hybrid towers. It collects structural response data from key components of the wind turbine hybrid tower and simultaneously incorporates operating condition data such as wind speed and power to construct a multi-channel time-series sample that fuses structural and operating condition features. Based on an unsupervised deep learning model, the fused features are modeled to learn the typical spatiotemporal response patterns of the wind turbine hybrid tower in a healthy state. The degree of structural health deviation is characterized by the reconstruction error between the input data and the model-reconstructed data. Combined with an adaptive threshold mechanism based on error statistical distribution, the method achieves anomaly determination and early warning for the structural health status of wind turbine hybrid towers. This invention does not rely on fault sample labels, effectively reducing the interference of operating condition changes on anomaly identification results, and enabling online monitoring of early stiffness degradation in wind turbine hybrid tower structures. It is suitable for long-term health monitoring scenarios of wind turbine hybrid towers.
Need to check novelty before this filing date? Find Prior Art