Improved spatiotemporal intrinsic mode decomposition method and its application in gear fault diagnosis

Abstract

Multi-channel vibration signal decomposition can fully exploit the interrelations among different sensor signals and effectively extract features of gear fault vibration signals and has been widely used. However, under complex and noisy environments, these methods still face issues such as insufficient robustness and limited decomposition performance. Spatiotemporal intrinsic mode decomposition (STIMD) separates spatiotemporal data, effectively solving the issue of blind source separation (BSS) for non-stationary spatiotemporal signals, thus achieving efficient separation of multi-channel signals. However, STIMD has two main drawbacks: (1) The decomposition results are highly sensitive to the selection of initial phases, and improper estimation of initial phases significantly impacts decomposition accuracy. (2) The method cannot autonomously sort the resulting components, which affects subsequent diagnostic analysis. To overcome these challenges, this study proposes an improved STIMD (ISTIMD) method. This approach uses the maximal overlap discrete wavelet packet transform to estimate the initial phases, which enables accurate separation of component signals. Additionally, considering the periodic and sparse characteristics of fault signals, the periodic sparsity index is defined to effectively sort the decomposed components. Enveloping spectrum analysis is then performed on the sorted components to reveal gear fault diagnosis. The experimental results show that, compared to other classic BSS methods, the proposed ISTIMD method demonstrates better performance in BSS of gear fault vibration signals with strong noise and accurate extraction of fault information, which provides a new approach for more accurate feature extraction and gear fault diagnosis.