Completely adaptive projection multivariate local characteristic-scale decomposition and its application to gear fault diagnosis

Abstract

Multichannel signal can be quickly and effectively decomposed using multivariate local characteristic-scale decomposition (MLCD). However, multi-channel signals acquired by sensors often have power imbalance characteristic, and MLCD uses uniform projection to estimate the baseline, which leads to inaccuracy of MLCD in decomposing such signals. Additionally, uniform projection introduces some projection vectors that can't precisely describe the multichannel signal, based on which, completely adaptive projection (CAP) is proposed in this paper. Completely adaptive projection multivariate local characteristic-scale decomposition (CAPMLCD) is developed on the basis of CAP. The simulation and experimental results show that CAPMLCD and MLCD perform better in terms of suppressing mode mixing, decomposition efficiency, and decomposition accuracy than multivariate empirical mode decomposition (MEMD) and adaptive projection intrinsically transformed multivariate empirical mode decomposition (APIT-MEMD). By comparing with MEMD, APIT-MEMD and MLCD, CAPMLCD has better decomposition accuracy and lower dependence on the number of projections.