Bearing Fault Diagnosis Based on AVMD and HPO-DBN

To overcome difficulties such as non-stationary vibrations, highdimensional feature redundancy, and mode selection issues that may arise during signal decomposition in bearing fault diagnosis. We propose an adaptive method called Adaptive Variational Mode Decomposition (AVMD) for extracting time-frequency domain characteristics from the bearing vibration displacement signals to the maximum extent possible. Next, the ReliefF algorithm is employed to select desired features, and an autoencoder is used to reduce the selected features dimensionally. Furthermore, because the Hunter-Prey Optimisation (HPO) algorithm can balance multiple objectives during the search process by utilising the concepts of hunter and prey to generate a better solution set, incorporating this algorithm into the Deep Belief Network (DBN) establishes an HPO-DBN fault diagnosis model. Subsequently, we validate the proposed method using both public datasets and field compressor data. Moreover, we compare the results with those obtained from the Support Vector Machine (SVM). The findings indicate that this approach enhances the bearing fault identification rate, thus supporting predictive maintenance of bearings.OPEN ACCESS Received: 13/08/2025 Accepted: 16/10/2025 Published: 23/01/2026