風力發電系統機械組件異常分析與檢測

由於綠色能源的發展，全球對於風力發電需求已快速上升，且在風力發電機裝置容量逐年提升情況下，有必要針對風機運轉狀況進行檢測，以避免因風機損壞對於電力系統之衝擊。而本研究將針對風機系統常見之齒輪箱漏油、軸心偏移及軸承損壞等3種異常情況進行分析與檢測。首先，為改進希爾伯-黃轉換端點效應對於訊號分析之影響，本研究採用兩種類神經網路模型，分別為倒傳遞類神經網路及改良式廣義回歸類神經網路，並評估這兩種模型對於端點效應改善成效。其次，應用5種分析法，分別為：1) 小波多重解析度分析、2) S轉換、3) 希爾伯-黃轉換、4) 基於端點效應處理之希爾伯-黃轉換及5) 包絡譜分析法，以分析發電機輸出電流訊號，再使用特徵擷取法擷取其特徵。最後，使用上述5種訊號分析法之特徵，並採用5種分類演算法進行風機異常檢測，分別為：1) 自適應徑向基底類神經網路、2) K平均值徑向基底類神經網路、3) 傳統徑向基底類神經網路、4) 倒傳遞類神經網路及5) K鄰近分類法。對於各種訊號分析法及分類演算法在評估比較方面，若訊號分析法採用小波多重解析度分析，且分類演算法採用自適應徑向基底類神經網路，其結果顯示，雖然運算時間大幅縮短，但抗雜訊能力較差；另外，若訊號分析法及分類演算法採用S轉換與K平均徑向基底類神經網路，則具較高之分類準確率，且在雜訊比為25dB時，仍有84.4%之分類準確率，具有較佳之抗雜訊能力。

關鍵字

旋轉機械；訊號分析；希爾伯-黃轉換；分類演算法

並列摘要

Due to the prosperous development of green power, the demand of wind energy booms. Meanwhile, the installation capacity of wind turbines become larger and larger, the diagnosis of wind turbine operation is essential to avoid the impact of wind turbine damages on power system. This paper aims to analyze and diagnose three common mechanical abnormalities of wind power system, including: gearbox lubrication leakage, rotor angular misalignment and damaged bearing. First, this paper proposes two artificial neural network models to improve the end effects of Hilbert Huang transform (HHT) on signal analysis, namely back propagation neural network (BPNN) and enhanced general regression neural network (EGRNN). Then, their performances will be evaluated. Second, this paper analyzes the output current of wind generator by: individually using: 1) Wavelet multi-resolution (Wavelet MRA), 2) S-transform, 3) HHT, 4) HHT based on end effect processing and 5) envelope analysis and obtains their features by feature extraction. Finally, this paper diagnoses errors by applying the five features and employing five classifiers, namely adaptive cluster radial basis neural network (ACRBFN), K-means radial basis neural network (KMRBFN), conventional radial basis neural network (CRBFN), BPNN and K-nearest neighbor algorithm (KNN). By comparison, the Wavelet MRA with ACRBFN needs less computing time but has worse noise tolerance. Moreover, the S-transform with KMRBFN has best classification accuracy and noise tolerance. And even with SNR=20dB noise, the accuracy still reaches 84.4%, which shows its better noise tolerance.

並列關鍵字

signal analysis ； rotating machine ； classifier ； hilbert-huang transform

參考文獻

[5] P. J. Tavner, J. Xiang, and F. Spinato, “Reliability analysis for wind turbines,” Wind Energy, vol. 10, pp. 1–18, 2007.

[7] P. D. McFadden, “Low frequency vibration generated by gear tooth impacts,” NDT International Elsevier, vol. 18, no. 5, pp. 279–282, 1985.

[8] S. T. Wan, H. M. Li, Y. G. Li, and F. C. Meng, “Analysis of stator winding parallel-connected branches circulating current and its application in generator fault diagnosis,” in Proceedings of the Conference Record. IEEE IAS Annual Meeting, 2005, vol. 1, pp 42-45.

[9] W. Yang, P. J. Tavner, and M. R. Wilkinson, “Condition monitoring and fault diagnosis of a wind turbine synchronous generator drive train,” in Proceedings of the IET Renewable Power Generation, 2009, vol. 3, No. 1, pp. 1-11.

[11] J. R. Cameron, W. T. Thomson, and S. Roach, “Vibration and current monitoring for detecting airgap eccentricity in large induction motors,” in Proceedings of the IEE Electric Engineering Applications, vol. 133, no. 3, pp. 155–163, May. 1986.

國際替代計量

風力發電系統機械組件異常分析與檢測

全文下載

主題瀏覽