Abstract
In this paper, a novel fault diagnosis method based on vibration signal analysis is proposed for fault diagnosis of bearings and gears. Firstly, the ensemble empirical mode decomposition (EEMD) is used to decompose the vibration signal into several subsequences, and a multi-entropy (ME) is proposed to make up the fusion features of the vibration signal. Secondly, an improved manifold learning algorithm, local and global preserving embedding (LGPE), is applied to compress the high-dimensional fusion feature set into a two-dimension feature set. Finally, according to the clustering accuracy of different feature set, the fault classification and diagnosis can be performed in the reduced two-dimension space. The performance of the proposed technique is tested on the fault of wind turbine transmission system. The application results indicate that the proposed method can achieve high accuracy of fault diagnosis.
What Problem Does This Paper Address?
Wind turbines are critical assets for clean energy, but they operate in harsh, remote environments that lead to significant mechanical wear. Among the various components, the transmission system—specifically gearboxes and bearings—represents a major point of failure. As shown in historical maintenance data, gearboxes consistently exhibit high failure rates compared to other components like generators or control systems, leading to expensive downtime and complex repairs.
Engineers typically use vibration sensors to monitor these systems, but the resulting signals are non-stationary, non-linear, and often drowned out by background noise. Traditional diagnostic methods struggle with two main issues:
- Feature Overload: Extracting too many mathematical features leads to the "curse of dimensionality," where redundant data makes classification harder.
- Structural Loss: Existing dimensionality reduction methods usually focus on either the "big picture" (global structure) or the "fine details" (local neighborhoods), but rarely both effectively.
Methods: The EEMD-ME-LGPE Framework
The authors propose a robust three-stage pipeline to overcome these challenges, integrating signal decomposition with advanced manifold learning:
- Signal Decomposition (EEMD): Ensemble Empirical Mode Decomposition breaks complex vibration signals into simpler components called Intrinsic Mode Functions (IMFs). By adding white noise during decomposition, EEMD prevents "mode mixing," ensuring the resulting signals are physically meaningful.
- Feature Fusion (Multi-entropy): Instead of a single metric, the system calculates three types of entropy—envelope, permutation, and energy entropy—from the most sensitive IMFs. This "Multi-entropy" (ME) approach captures the complexity, sparseness, and energy distribution of the signal simultaneously.
- Dimensionality Reduction (LGPE): The core innovation is Local and Global Preserving Embedding (LGPE). This manifold learning algorithm compresses high-dimensional features into a 2D map by minimizing local neighbor distances while maximizing global variance. It also includes an orthogonal constraint to specifically filter out noise during this compression.
Datasets and Validation
The researchers validated their model using two primary sources:
- CWRU Bearing Data: The industry-standard benchmark from the Case Western Reserve University Bearing Data Center, covering normal states and various bearing faults.
- QPZZ-II Test Rig: A specialized laboratory setup used to simulate gear-specific failures, including normal operation, gear wear, and broken teeth.
Key Results and Insights
The proposed LGPE method was compared against mainstream algorithms like Kernel Principal Component Analysis (KPCA) and Locality Preserving Projection (LPP).
- Clustering Performance: LGPE produced perfectly separated clusters in 2D space. In contrast, traditional methods showed significant overlap between different fault types (e.g., mixing up "outer race faults" with "ball faults"), which would lead to false diagnoses in the field.
- Accuracy: The EEMD-ME-LGPE model achieved 100% accuracy in identifying the recorded fault classes for both the bearing and gear datasets.
The paper demonstrates that mechanical wear shows up clearly as changes in signal randomness or complexity (entropy). By preserving both the local nonlinear manifold and global variance, LGPE creates clear visual "fingerprints" of machine health, making automated classification highly reliable.
Limitations and Future Work
While the results are impressive, the authors note areas for further development:
- Variable Conditions: Future tests should address shifting wind speeds and alternating loads typical of real-world environments.
- Real-time Implementation: Further work is needed to optimize the computational speed of EEMD and LGPE for real-time, edge-computing applications directly on the turbine.
Summary updated