A nonlinear dynamics method using multi-sensor signal fusion for fault diagnosis of rotating machinery

Abstract

Deep mining abnormal information from operation data is a crucial step in fault diagnosis of equipment, and it holds significant importance for ensuring the efficient operation of rotating machinery. The nonlinear dynamics methods represented by multivariate multiscale entropy have shown good application effects in quantifying the fault characteristics of rotating machinery using multiple sensor signals. However, these methods essentially belong to the category of data-level fusion, which suffers from drawbacks such as poor real-time performance, limited capability to handle only similar types of sensors, and significant influence from sensor information. This paper develops a novel tool named enhanced hierarchical Poincaré plot index (EHPPI), for extracting anomaly information from multi-source signals via feature-level fusion. Firstly, the Poincaré plot index is extended to create the EHPPI, allowing for the extraction of information from signals at various frequency scales. Subsequently, EHHPI is utilized to extract information from all channel signals. Ultimately, we concatenate the information extracted from all channels by EHPPI to form features and integrate them with random forests to identify faults in rotating machinery. The EHPPI and other popular nonlinear dynamics metrics are applied in different scenarios, such as simulation faults, experimental bench faults, and real machine faults, whose results strongly prove its advantages. The EHPPI has a favorable effect on improving the operational efficiency of rotating machinery.