Rolling bearings play a critical role in rotating machinery, with their fatigue
life directly impacting equipment’s operational reliability. This underscores
the significant engineering application value of “fault diagnosis” (FD)
technology for rolling bearings in mechanical, automation, and aerospace
domains. Literature reviews highlight that a substantial portion of failures in
machinery such as jet turbine engines, wind turbines, gear reducers, and
induction machines are attributable to bearing issues. Early fault detection and
preventive maintenance are therefore imperative for ensuring the smooth
operation of rotating machinery. This paper focuses on rolling bearings, delving
deep into FD technology using machine learning principles. It analyses the
structure and common failure modes of rolling bearings, discussing an FD method
based on machine learning. Specifically, the SE-DRN (“squeeze-exclusion deep
residual network”) approach is employed, leveraging “variational modal
decomposition” (VMD) to decompose bearing vibration signals and reorganize the
resulting “intrinsic mode function” (IMF) components into an IMF component
signal matrix. This matrix is then processed by a depth residual network with a
channel attention mechanism for feature extraction and recognition, forming the
SE-DRN-based FD model for rolling bearings. The research attains a remarkable
average diagnostic accuracy of 98% across five different bearing state types,
underscoring its superior accuracy compared to existing literature, thus
showcasing the effectiveness of the SE-DRN approach in rolling bearing FD
technology.