Hierarchical bandwidth-adaptive variational mode decomposition: algorithms and applications

Area of Science:

• Signal Processing
• Machine Learning
• Mechanical Engineering

Background:

• Nonlinear and non-stationary signal analysis faces challenges like mode mixing and limited parameter adaptivity.
• Existing methods often require predefining the number of intrinsic modes, limiting flexibility.

Purpose of the Study:

• To introduce a hierarchical bandwidth-adaptive variational mode decomposition (HBVMD) method.
• To address mode mixing and improve parameter adaptivity in signal decomposition.
• To enhance the robustness and interpretability of industrial fault diagnosis.

Main Methods:

• HBVMD employs a top-down binary recursive strategy for signal decomposition into low- and high-frequency components.
• A time-varying mutual information criterion and modal energy threshold determine recursion termination and decomposition depth.
• A dynamic dual-bandwidth penalty scheme and adaptive augmented Lagrangian step size are utilized.

Main Results:

• HBVMD effectively decomposes signals without requiring a predefined number of modes.
• Experiments on bearing fault datasets demonstrate high diagnostic accuracy and reliability.
• The method excels in composite fault detection and feature extraction, outperforming existing approaches.

Conclusions:

• HBVMD offers a robust and interpretable framework for industrial fault diagnosis.
• The adaptive nature of HBVMD overcomes limitations of traditional signal decomposition techniques.
• This method significantly improves the analysis of complex nonlinear and non-stationary signals.