Bearing Fault Online Identification Based on ANFIS

Abstract

Effectiveness of online bearing status monitoring (OBSM) depends deeply on the online data processing ability and the sensitivity of data features used to recognize the mechanical-system dynamic response change. Focusing on these, we present a novel method of OBSM based on singular spectrum analysis (SSA) and adaptive neuro-fuzzy inference system (ANFIS) with the highlights as follows. A sensitive and stable multi-feature is discovered to better the ability to distill the valuable information in noisy and massive databases (NMDs) and process impulse-noise in them. The SSA-based high-frequency noise removal solution, the ANFIS’ interpolating and identifying capability, and the dual function of the proposed multi-feature are combined in a new algorithm named AfOBSM for building a system of OBSM through two phases, offline and online. The offline is to identify the mechanical-system in the presence of the typical kinds of bearing faults. The ANFIS is trained in this phase using a training dataset. Meanwhile, the online is to estimate online the real status of the bearing(s) based on the trained ANFIS and a monitoring dataset. Surveys from an experimental-system were performed. The obtained results showed the positive effects of the AfOBSM.