Bearing Fault Diagnosis Based on Measured Data Online Processing, Domain Fusion, and ANFIS

Processing noise online in sensors-based measurement data (SMD) and mitigating the effect of domain drift are always challenges. As a result, it negatively impacts the effectiveness and feasibility of data-driven model (DDM)-based mechanical-system fault identification (MFI). Here, we propose an online bearing fault diagnosis method named ANFIS-BFDM by using an adaptive neurofuzzy inference system (ANFIS). Reduction in the influence of domain drift between the source domain and target domain (DDSTD) is considered in both the data processing and fault identification. Online solutions for preprocessing SMD and exploiting the filtered data to label the target domain are presented in a fusion domain deriving from the source and target domains. First, in the offline phase, frequency-based splitting of SMD into different time series is performed to cancel the high-frequency region. An optimal data screening threshold (ODST) is distilled in the remaining low-frequency data to develop an impulse noise filter named FIN. An ANFIS then identifies the dynamic response of the bearing(s) via the filtered data. The FIN and ANFIS are finally exploited during the online phase to filter noise and recognize the object's health status online. The survey results reflect the positive effects of the method, even if severe impulse noise appears in the databases.