Composite Multiscale Transition Permutation Entropy-Based Fault Diagnosis of Bearings

Jing Guo, Biao Ma, Tiangang Zou, Lin Gui, Yongbo LiSensors 2022 · DOI

TL;DR

This paper introduces Composite Multiscale Transition Permutation Entropy (CMTPE), a novel method for analyzing vibration signals to diagnose bearing faults. CMTPE improves upon existing multiscale entropy techniques by addressing information loss during signal processing, leading to more accurate and robust fault detection, especially in noisy conditions.

What Problem Does This Paper Address?

Rotating machinery, crucial in industries like aerospace and automotive, often suffers from bearing failures, which can lead to costly downtime and safety hazards. Accurately diagnosing these faults early is essential for predictive maintenance. Vibration signal analysis is a common technique, but these signals are often nonlinear, non-stationary, and susceptible to noise, making it challenging to extract reliable fault features. Existing entropy-based methods, while useful for quantifying signal complexity, often struggle with single-scale analysis or suffer from information loss and reduced accuracy when applied across multiple scales. This paper aims to overcome these limitations by proposing a more comprehensive and robust multiscale entropy approach for bearing fault diagnosis.

Methods

The core of this paper is the development of Composite Multiscale Transition Permutation Entropy (CMTPE). Here's a breakdown of the methodology:

  • Transition Permutation Entropy (TPE): This is the foundational entropy measure. It quantifies signal complexity by analyzing the transition probabilities between ordinal patterns (the relative order of values) within a time series. TPE uses the positive eigenvalues of the transition probability matrix for its calculation.
  • Multiscale Transition Permutation Entropy (MTPE): To address the limitations of single-scale analysis, MTPE extends TPE by applying it to coarse-grained versions of the original time series at various scales. This allows for the analysis of signal dynamics at different resolutions.
  • Composite Multiscale Transition Permutation Entropy (CMTPE): This is the novel contribution. Traditional multiscale methods (like MTPE) often only consider the first coarse-grained time series at each scale. CMTPE, however, generates all possible coarse-grained time series for a given scale factor by using different starting points. The final CMTPE value is the average of the TPE values calculated from all these composite coarse-grained time series. This approach aims to retain more information from the original signal and improve the accuracy and stability of entropy estimation.
  • Fault Diagnosis Strategy: The extracted CMTPE features are then fed into an Extreme Learning Machine (ELM) classifier. This is a supervised learning approach where the ELM is trained on a subset of the features and then used to predict the fault type for the remaining test data.

Datasets

The paper utilizes two types of datasets for evaluation:

  • Simulated Bearing Signals: Three types of simulated bearing faults were generated: outer race fault, inner race fault, and ball fault. These signals were intentionally corrupted with Gaussian white noise at various Signal-to-Noise Ratios (SNRs) ranging from 10 dB to 40 dB to test the robustness of the methods.
  • Experimental Bearing Data: Experimental data was collected from a bearing fault test rig. Five health conditions were tested: normal, inner race crack 4mm (IRC), outer race crack 4 mm (ORC), inner race pitting 3 mm (IRP), and outer race pitting 3 mm (ORP). Further experiments were conducted with minor faults (different degrees of crack and pitting faults) to assess the method's sensitivity.

Key Insights and Ideas

  • The "Composite" Advantage: The core insight is that by considering all coarse-grained time series at each scale (rather than just the first), CMTPE captures a richer representation of the signal's complexity and dynamics. This "composite" approach mitigates the information loss that can occur with simpler multiscale methods.
  • Improved Robustness to Noise: CMTPE demonstrates superior performance compared to TPE, MTPE, and MPE, especially under low SNR conditions. This suggests that its comprehensive feature extraction is less susceptible to noise interference.
  • Enhanced Feature Extraction for Minor Faults: The study highlights CMTPE's effectiveness in identifying even subtle faults (e.g., minor cracks and pitting). This is crucial for early detection and preventing catastrophic failures.
  • Stability Matters: Beyond accuracy, CMTPE exhibits lower variance in its results across multiple runs, indicating greater stability and reliability in its diagnostic capabilities.

Key Results

  • Simulation Results:
    • CMTPE consistently achieved higher test accuracy than TPE, MTPE, and MPE across all tested SNR values.
  • Even at an SNR of 20 dB, CMTPE maintained 100% accuracy, while MTPE and MPE dropped significantly.
  • At an SNR of 10 dB, CMTPE achieved 89.60% accuracy, outperforming MTPE (61.90%) and MPE (71.73%).
  • Experimental Results:
    • CMTPE achieved the highest average test accuracy of 98.60% among the compared methods.
  • CMTPE also had the lowest variance (0.65%), indicating superior stability.
  • Confusion matrices showed CMTPE had only four misclassifications, achieving 98.4% accuracy, the highest among all methods.
  • For minor fault identification (crack and pitting), CMTPE achieved accuracies of 96.46% and 99.67%, respectively, significantly outperforming other methods, especially TPE.
  • Visualization: t-SNE visualization of extracted features showed that CMTPE produced the largest inter-cluster distances and smallest intra-cluster distances, indicating the most distinct and well-separated feature representations.

Limitations and Future Work

  • Noise Reduction: The paper acknowledges that in cases of very low or negative SNR, filtering and other noise reduction methods might be necessary to preprocess the signal before applying CMTPE.
  • Future Work: The authors suggest exploring the effectiveness of combining CMTPE with other noise reduction techniques in future research. They also plan to test the method's performance with different types of noise and more complex fault scenarios.

Summary updated

Abstract

When considering the transition probability matrix of ordinal patterns, transition permutation entropy (TPE) can effectively extract fault features by quantifying the irregularity and complexity of signals. However, TPE can only characterize the complexity of the vibration signals at a single scale. Therefore, a multiscale transition permutation entropy (MTPE) technique has been proposed. However, the original multiscale method still has some inherent defects in the coarse-grained process, such as considerably shortening the length of time series at large scale, which leads to a low entropy evaluation accuracy. In order to solve these problems, a composite multiscale transition permutation entropy (CMTPE) method was proposed in order to improve the incomplete coarse-grained analysis of MTPE by avoiding the loss of some key information in the original fault signals, and to improve the performance of feature extraction, robustness to noise, and accuracy of entropy estimation. A fault diagnosis strategy based on CMTPE and an extreme learning machine (ELM) was proposed. Both simulation and experimental signals verified the advantages of the proposed CMTPE method. The results show that, compared with other comparison strategies, this strategy has better robustness, and can carry out feature recognition and bearing fault diagnosis more accurately and with improved stability.
BibTeX
@inproceedings{guo2022composite,
  author = {Jing Guo and Biao Ma and Tiangang Zou and Lin Gui and Yongbo Li},
  title = {{Composite Multiscale Transition Permutation Entropy-Based Fault Diagnosis of Bearings}},
  booktitle = {Sensors},
  year = {2022},
  doi = {10.3390/s22207809},
}
Figure