关键词: 振动信号分析, 故障诊断, IENEMD, FastICA, 盲源分离, 包络谱分析

Abstract: To solve the problem of unsteady, nonlinear and under-determined blind source separation (BSS) of single-channel vibration signals of rotating machinery in strong noise background, this paper presents a combined analysis framework based on improved ensemble noise-reconstructed empirical mode decomposition(IENEMD) and fast independent component analysis(FastICA). The method uses IENEMD to adaptively extract the intrinsic noise of the signal, avoid introducing external Gaussian white noise, effectively suppresses mode mixing and generates high-quality virtual multi-channel signals. Then, FastICA is used to realize the blind source separation of nonlinear mixed signals, and the dual criterion of the gradient and Pearson correlation coefficient(PCC) is used to screen the fault component, and finally the accurate identification of the fault frequency is realized through the envelope spectrum analysis. Based on Case Western Reserve University bearing dataset, the experiments show that this method has good adaptability, robustness and calculation efficiency in nonlinear and non-stationary signal processing, and the fault frequency identification error is less than 0.5%. Compared with CEEMDAN-FasteICA and VMD-FastICA, this method achieves a better balance between feature extraction accuracy and conputational efficiency.

Key words: vibration signal analysis, fault diagnosis, IENEMD, FastICA, blind source separation, envelope spectrum analysis