Abstract:
To effectively eliminate external noise from the vibration signals of dense medium cyclones and enhance the efficiency and reliability of fault detection, a novel signal processing method combining Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Sample Entropy (SampEn), and an improved wavelet threshold technique is proposed. This method addresses the challenges of poor noise reduction and difficulty in feature extraction from dense medium cyclone vibration signals. Firstly, CEEMDAN is employed to suppress mode mixing and generate Intrinsic Mode Functions (IMFs). Subsequently, Sample Entropy is used to identify and filter noise components with higher entropy values. The selected noisy components are then processed using an improved wavelet threshold to remove noise while preserving effective signal features. Finally, the denoised IMF components are reconstructed along with the noise-free modal components to obtain an optimized signal. Simulation experiments demonstrate that the proposed method significantly outperforms traditional wavelet thresholding, CEEMDAN, and the combined CEEMDAN-Wavelet threshold method in terms of Signal-to-Noise Ratio (
SNR), Mean Squared Error (
MSE), and Normalized Cross-Correlation Coefficient (
NCC). Specifically, the
SNR increases to 33.36 dB, the
MSE decreases to 0.009 5, and the
NCC reaches 0.999 9. In a practical validation, the vibration signal processing results from the Dense Medium Cyclone at the Wobei Coal Preparation Plant indicate that the first six IMF components, as identified by Sample Entropy, are predominantly noise. The proposed method effectively removes high-frequency noise while retaining low-frequency fault characteristics, demonstrating superior denoising performance and feature preservation in practical applications. Additionally, the Lempel-Ziv Complexity (
LZC) is restored to 47. The proposed CEEM-DAN-SampEn-Improved Wavelet Threshold method achieves efficient noise reduction of dense medium cyclone vibration signals through multi-scale decomposition, entropy-based filtering, and optimized threshold denoising. It provides clearer signals for fault diagnosis and has demonstrated excellent performance in both simulation and practical validation, highlighting its significant theoretical value and engineering application potential.