An Anomaly Detection Method for Sliding Bearings Under Electromagnetic Interference Based on an Asymmetric Autoencoder and a Dual-Indicator Diagnostic Plane

As core components of marine power systems, sliding bearings directly affect the operational safety of ships. To address the challenges that early weak anomaly features of sliding bearings are easily masked by electromagnetic interference (EMI) and that a single evaluation indicator may respond differently across wear conditions, this paper proposes an asymmetric autoencoder dual-indicator anomaly detection (AADI-AD) method. First, adaptive notch filtering and high-pass filtering are employed to suppress EMI. Then, horizontal and vertical asymmetric convolution kernels are introduced, and an autoencoder is constructed in combination with a dynamic feature allocation mechanism to represent directional time–frequency structures. Finally, a dual-indicator diagnostic plane is established by combining time-domain kurtosis with the mean squared error (MSE) for time–frequency image reconstruction. The experimental results show that the two indicators provide complementary responses across the examined wear conditions and reduce missed detections compared with single-MSE evaluation. These results demonstrate the effectiveness of the proposed method for sliding-bearing anomaly detection under the EMI-contaminated conditions represented by the present experimental platform.