多传感器数据融合的风电齿轮箱性能衰退评估

Aiming at the problem that wind turbine generator gearbox was difficult to extract the effective features of vibration signals for performance degradation analysis caused by the complex power driving structure and terrible working conditions, a method based on multi-sensor information fusion was proposed. Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Kernel Principal Component Analysis (KPCA) and Hotelling's T-squared statistic (T²) were used to achieve the assessment of performance degradation. A CEEMDAN-KPCA based method was applied to denoise and reconstruct the nonlinear and unstable vibration signals of the life cycle; the reconstructed signals were fused with KPCA, and then continuous T² (C-T²) and related time domain features were extracted to establish the performance degradation models. Experimental results showed that the proposed method had a remarkable effect on denoising the vibration signals, and C-T² was effective to solve the expansion of feature dimension caused by multiple sets of vibration signals. As well as, models of related time domain features by C-T² could evaluate the performance degradation more accurate than the time-domain features of vibration signal.