Rolling bearing fault diagnosis based on SSA optimized self-adaptive DBN

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Rolling bearing vibration signals exhibit complex non-stationary and time-varying characteristics.
• Accurate identification of bearing faults from these signals is challenging due to data complexity.
• Existing methods struggle with the inherent variability in operational environments.

Purpose of the Study:

• To propose an innovative optimized adaptive deep belief network (SADBN) for rolling bearing fault identification.
• To enhance the classification accuracy of deep belief networks (DBNs) for bearing fault diagnosis.
• To leverage intelligent optimization algorithms for improved DBN performance.

Main Methods:

• A deep belief network (DBN) was pre-trained using minimum batch stochastic gradient descent.
• Backpropagation neural network and conjugate gradient descent were employed for supervised fine-tuning.
• The Salp Swarm Algorithm was utilized for optimizing the DBN architecture and parameters.

Main Results:

• The proposed SADBN method demonstrated significant improvements in classification accuracy for bearing faults.
• Simulations using experimental data validated the effectiveness of the optimized deep learning approach.
• The study successfully addressed the challenges of non-stationary and time-varying vibration signals.

Conclusions:

• The optimized adaptive deep belief network (SADBN) offers a robust solution for rolling bearing fault identification.
• The integration of intelligent optimization algorithms enhances deep learning model performance in mechanical diagnostics.
• This approach provides a reliable method for analyzing complex vibration data in industrial applications.