Integrated Method for Shaft Bearing Fault Diagnosis Based on VMD Signal Reconstruction and Enhanced YOLOv8 Model

To address the weak generalisation and anti-interference capability of the current intelligent detection model of the shaft system, this paper innovatively proposes a method for diagnosing faults in vessel propulsion shaft bearings. It first utilizes Variational Mode Decomposition (VMD) to reconstruct collected vibration signals in order to reduce the impact of noise. Subsequently, it encodes the time series data into two-dimensional images using Gramian Angular Field (GAF) to retain temporal information in the data and reduce computational complexity. These images are then fed into a Yolov8 network model integrated with a Convolutional Block Attention Module (CBAM). The experiments involve training the model with constant speed fault data to classify variable speed fault data, assessing the model's cross domain analysis capabilities, and testing its noise resistance by adding noise to the original vibration signals. The experimental results demonstrate that this method can successfully identify variable speed vibration signals that have not been trained on as well as signals with added noise. Consequently, this approach addresses the issue of decreased diagnostic accuracy due to noise interference when using vibration signals for fault diagnosis, along with the challenge of adapting the model to different operating conditions.