Design and development of a time-frequency analysis system for vibration signals

Abstract

In this study, a set of time-frequency analysis system integrating short-time Fourier transform (STFT), wavelet transform (WT) and Hilbert-Huang transform (HHT) is designed and developed for the problem that traditional time-frequency analysis methods are difficult to capture the dynamic features of complex vibration signals in the diagnosis of rotating machinery faults; the system adopts the architecture of “Data Acquisition - 5-layered db8 Wavelet Preprocessing - Multi-Algorithmic The system adopts the architecture of “data acquisition - 5-layer db8 wavelet preprocessing - multi-algorithm fusion analysis - machine learning diagnosis - visualization output”, in which the STFT achieves time-frequency localization through Hann window 50% overlap optimization, the WT performs multi-scale decomposition based on the Morlet wavelet to capture transient features, and the HHT deals with nonlinear signals through empirical modal decomposition (EMD) coupled with end-point corrections, and the HHT is applied to the signal by using the Python SciPy/NumPy library of Python is used to accelerate the parallel computation, which improves the STFT computation efficiency by 2.34 times and reduces the memory consumption by 53% compared with MATLAB; the system achieves 100% fault detection rate after validation of the CWRU bearings and SpectraQuest gearbox dataset, among which the HHT achieves 98.7% adaptability to the nonlinear faults, and realizes 100% differentiation of health/fault states by PCA reduction, and visualization of the HHT and the HHT. / The HHT is 98.7% adaptive to nonlinear faults, and 100% differentiation between healthy and faulty states is achieved through PCA dimension reduction, and the visualization module supports time-frequency spectrograms to visually present the fault characteristics, which provides a feasible engineering solution for real-time monitoring and accurate diagnosis of industrial rotating machinery.