103 / 2021-07-22 04:29:53

A Subspace Domain Adaptation Method: SSA-Theoretic Drift Correction for Gear Fault Diagnosis under Varying Working Conditions

To boost the performance of gear fault diagnosis (GFD) under various working conditions, a new subspace domain adaptation (DA) strategy is proposed in this paper. Here the source domain comes from historical data or meshing parts, and the target domain data is collected at current moment or under different working conditions. Based on previous DA strategy that learns domain-invariant characteristics between domains, this paper integrates stationary subspace analysis (SSA) with DA to achieve performance improvement further. The SSA first decomposes the vibration signal into stationary and nonstationary subspaces. Subsequently, the nonstationary signal the maximum kurtosis is chosen as major fault component, and taken as inputs into maximum mean discrepancy (MMD)-based DA model to correct the feature drift. Related experiments prove that the proposed method can actualize the coherence of feature distribution and improve GFD performance ulteriorly.