The early faults of sliding bearings are highly concealed. To accurately predict their vibration amplitude

a deep learning model incorporating a YOLOv8-optimized CBAM attention mechanism is proposed.The CBAM module is embedded between the Backbone and Neck to enhance the model's focus on critical vibration features. Additionally

an improved complete intersection over union loss function is employed to enhance object detection accuracy. Considering the nonlinear and non-stationary characteristics of vibration data

the empirical mode decomposition(EMD) method is integrated into the model to improve the accuracy of vibration state prediction. The experimental results show that

on the 600 MW steam turbine operation dataset

this method improves the detection accuracy by 2.85 percentage points and 8.50 percentage points compared with that of the conventional YOLOv8 and YOLOv7

respectively. Moreover

the root mean square error(RMSE) is reduces

and the mean absolute error(MAE) decreases. Furthermore

in high-noise environments

the model's error fluctuation reduces by 30% compared with that of the conventional methods

demonstrating stronger generalization ability and stability.