Weak fault signal extraction for gearbox based on parallelizable underdetermined blind separation

摘要

Timely detection of the degradation and defects is crucial to prevent the gearbox system's performance deteriorates to an unacceptable extent. However, the fault signal is extremely weak and easily affected by the environmental noise at the early stages of the failure. In this paper, we consider the problem of extracting the underlying useful weak fault signals from noisy measurements. To accomplish this task, we develop a Parallelizable Underdetermined Blind Separation (ParUBS) method based on Sparse Parallelizable Tensor Decompositions (ParCube) and Second-Order Blind Identification (SOBI). By modeling the observed signals as superposition of a set of source signals without the aid of prior information about the source signals and the mixing process, a blind source separation problem is obtained. We propose to solve this problem through a parallel tensor decomposition method which can be seen as a generalization to underdetermined mixtures of the well-known SOBI algorithm. The performance of the method was evaluated through a set of numerical experiments on synthetic datas. Results shows that the proposed method to be able to estimate the mixing matrix of underlying useful weak fault signals. Moreover, the ParUBS method can also enables us to improve computation efficiency and reduce the number of vibration sensors compared with the classical blind separation method.