TL;DR
This paper introduces a novel method for identifying crack-rubbing coupling faults in marine bearings operating under high water pressure. By employing polar symmetry mode decomposition and dimensionality reduction, the approach effectively extracts fault features and achieves high recognition accuracy, addressing a critical need for improved fault detection in demanding marine environments.
What Problem Does This Paper Address?
Marine bearings are crucial components for the safe and stable operation of ships. However, they are susceptible to various failure modes, including fatigue pitting, plastic deformation, and abrasive wear. A particularly challenging failure mode, especially under high water pressure, is the crack-rubbing coupling fault. Existing fault identification methods often struggle with the non-stationary and nonlinear nature of vibration signals generated by such faults, particularly in early detection stages. Furthermore, the complex multi-component failure states in real-world scenarios make it difficult to isolate and analyze individual fault characteristics. This paper aims to bridge this gap by proposing a robust method for accurately recognizing crack-rubbing coupling faults in marine bearings under high water pressure.
Methods
The proposed method integrates several key techniques:
- Polar Symmetry Mode Decomposition (ESMD): This is the core signal processing technique. It's presented as an improvement over Empirical Mode Decomposition (EMD) and Masked EMD (MEMD), aiming for better trend separation, abnormal diagnosis, and time-frequency analysis. ESMD is described as an adaptive, baseless decomposition mode suitable for adaptive fault identification. The paper outlines the ESMD algorithm, which involves identifying extreme points, interpolating to form curves, and iteratively subtracting these curves to extract Intrinsic Mode Functions (IMFs).
- Dimensionality Reduction: To handle the complexity of the vibration signals and improve efficiency and accuracy, a dimensionality reduction technique is employed. This method involves dividing the system into subsystems, identifying nonlinear relationships between them, and then using one subsystem to represent others. This process is guided by invariant manifold theory, which helps in understanding the system's dynamics and reducing its order. The paper mentions using concepts like central manifolds and nonlinear Galerkin methods as theoretical underpinnings.
- Feature Extraction: After dimensionality reduction, key features are extracted from the processed signal. These features are characterized from the time domain, energy, and entropy. Specific metrics like sample entropy and Kurtosis are mentioned for their ability to capture signal complexity and impact pulses, respectively, which are relevant for bearing fault detection.
- Probabilistic Neural Network (PNN) Classification: A PNN is used as the classifier to recognize the identified fault. PNNs are chosen for their simple learning and training process, quick achievement of training standards, accurate data classification, and robustness to noise. The PNN takes the extracted feature parameters as input to identify the bearing fault.
Datasets
The paper describes an experimental setup using a Matlab platform with a test device including a motor, torque sensor, and power meter. Bearing specifications for SKF6203 and SKF6205 models are provided in Table 1. For training and verification, 50 groups of data were randomly selected from each fault data group, with 150 training sets and 30 test sets used in total. The paper does not appear to introduce a new benchmark dataset but rather uses data generated from their experimental setup.
Key Insights and Ideas
- ESMD's Advantage: The paper posits that ESMD offers a more adaptive and effective decomposition than traditional EMD, particularly for complex, non-stationary signals like those from faulty bearings.
- Dimensionality Reduction for Clarity: Reducing the dimensionality of the system is shown to effectively suppress noise and highlight the characteristic signals of the crack-rubbing fault, making subsequent analysis more precise.
- Synergy of Techniques: The combination of ESMD for signal decomposition, dimensionality reduction for simplification, and PNN for classification creates a powerful pipeline for fault recognition.
- High Water Pressure Relevance: The method is specifically tailored and validated for high water pressure conditions, a critical factor for marine applications where such environments are common.
Key Results
The experimental results demonstrate the effectiveness of the proposed method:
- Dimensionality Reduction: Figure 7 visually shows the significant reduction in signal fluctuations and noise after dimensionality reduction, with 10 reductions leading to a well-suppressed characteristic signal.
- Fault Recognition Accuracy: The paper presents comparative results in Figure 8, showing that the proposed method achieves higher fault recognition accuracy across minor, moderate, and severe fault categories compared to methods from literature [4], [5], and [6]. For instance, for minor faults, the proposed method achieved 96% accuracy, significantly outperforming others. For moderate faults, it reached 95-96% accuracy, and for severe faults, it achieved 95-97% accuracy, consistently higher than the compared methods.
Limitations and Future Work
- Limitations: The paper acknowledges that the ESMD algorithm's definition of IMF differs from EMD, and the selection of the number of IMFs to use for characterization (denoted by 'n') depends on the specific vibration signal decomposition. The paper also mentions that the "false IMF caused by interference in the polar symmetry mode decomposition" was considered, implying this is a potential issue that needs to be managed.
- Future Work: The authors suggest that in the future, "online estimation of bearing life will be studied to effectively reduce the impact of sudden failures on related work." This indicates a move towards predictive maintenance rather than just fault detection.
Summary updated