Stochastic Resonance with Parameter Estimation for Enhancing Unknown Compound Fault Detection of Bearings

Min Xu, Chao Zheng, Kelei Sun, Li Xu, Zijian Qiao, Zhihui LaiSensors 2023 · OpenReview

TL;DR

This paper introduces a novel stochastic resonance (SR) method for detecting subtle bearing faults that doesn't require prior knowledge of the fault type or its characteristics. By adaptively estimating SR parameters directly from the signal, the method enhances weak fault signatures more effectively and efficiently than existing SR techniques and outperforms the widely used fast kurtogram method.

What Problem Does This Paper Address?

Detecting early-stage faults in rotating machinery, particularly bearings, is crucial for preventing catastrophic failures and minimizing downtime. Many existing fault detection methods, including those based on stochastic resonance (SR), struggle with "weak fault signatures" – faint signals buried in background noise. A significant limitation of current SR-based approaches is their reliance on prior knowledge of the fault characteristics to tune SR parameters. This often involves using metrics like signal-to-noise ratio (SNR), which can lead to false positives or missed detections if the assumed fault signature is incorrect or unknown. In real-world industrial settings, obtaining this prior knowledge is often impractical or impossible, hindering the effectiveness of these advanced diagnostic tools. This paper aims to bridge this gap by developing an SR method that can autonomously estimate its parameters, making it suitable for diagnosing unknown or complex bearing faults without needing pre-existing information.

Methods

The core of this paper is a novel "parameter-matched second-order SR method." Traditional second-order SR systems have mathematical conditions for optimal performance, often expressed as a relationship between system parameters (like damping factor and noise intensity) and the signal's characteristics. The authors leverage these conditions, specifically focusing on the relationship between the noise intensity ($D$) and other system parameters ($a$, $b$, $\gamma$, $f_0$).

The key innovation lies in how they handle these parameters:

  • Parameter Estimation: Instead of relying on prior knowledge, the method estimates the noise intensity ($D$) from the raw vibration signal using the Maximum Likelihood Estimation (MLE) principle.
  • Adaptive Tuning: The damping factor ($\gamma$) is then tuned within a mathematically derived range to maximize the Signal-to-Noise Ratio Gain (SNRG). This iterative tuning process, guided by the SR conditions, allows the system to adapt to the specific signal being analyzed.
  • Signal Pre-processing: A Hilbert envelope demodulation is applied to the raw vibration signal to extract the underlying fault signatures, which are often modulated by natural machine vibrations.
  • Second-Order SR System: The method utilizes a second-order nonlinear system with a bistable potential, which inherently possesses a band-pass filtering characteristic, making it suitable for enhancing weak periodic signals.
  • Normalized Transformation: To handle "large-parameter" real-world signals (where parameters might be outside the typical small-parameter SR regime), a normalized transformation is applied to the SR system equation, allowing the method to be applicable to a wider range of industrial signals.

Datasets

The study utilizes experimental data from a bearing accelerated degradation test. This test involved running a bearing until failure, acquiring vibration signals using accelerometers placed on the bearing housing. The data was collected at a sampling frequency of 25.6 kHz, with a rotating speed of 2100 rpm and a radial force of 12 kN. The paper does not introduce a new benchmark dataset but uses this experimental setup to validate the proposed method against real-world bearing fault conditions.

Key Insights and Ideas

  • Knowledge-Agnostic Fault Detection: The most significant insight is that effective SR-based fault detection can be achieved without prior knowledge of the fault signature. This dramatically expands the applicability of SR to real-world industrial scenarios where fault types are often unknown or complex.
  • Adaptive Parameter Tuning is Key: By adaptively estimating noise intensity and tuning the damping factor to maximize SNRG, the method effectively "matches" the SR system to the signal, leading to superior enhancement of weak fault signatures.
  • Mathematical Conditions as Guides: The paper demonstrates that the theoretical conditions for triggering SR can be directly translated into a practical, adaptive parameter estimation and optimization framework.
  • Simplicity and Efficiency: Compared to other SR methods that require extensive parameter optimization using AI techniques, this approach is described as simpler and less time-consuming due to its direct parameter estimation and limited tuning range.

Key Results

The experimental verification on bearing fault data yielded promising results:

  • Enhanced Weak Fault Signatures: The proposed method successfully enhanced weak fault signatures, making them clearly visible in the frequency spectrum. Specifically, it was able to detect both inner race and outer race wear signatures, which were difficult to discern with traditional methods.
  • Superior to Fast Kurtogram: In a direct comparison, the proposed SR method outperformed the widely used fast kurtogram method. The fast kurtogram, when applied to the same dataset, failed to detect the weak fault characteristics of the tested bearing, whereas the proposed method successfully identified them.
  • Comparable Performance to Prior-Knowledge SR: The paper claims the proposed method achieves detection performance comparable to existing SR methods that do rely on prior knowledge, but with the significant advantage of not needing that knowledge.

Limitations and Future Work

  • Limitations:
    • The paper focuses on second-order SR systems. While effective, other SR orders or system types might offer different advantages.
  • The method's performance with extremely low signal-to-noise ratios or highly complex, multi-fault scenarios might require further investigation.
  • While described as simpler, the MLE estimation and iterative tuning still involve computational steps.
  • Future Work:
    • The authors suggest that the method's inherent mathematical limitations on the damping factor and noise intensity could be further explored for optimization.
  • Extending this parameter estimation approach to other types of SR systems (e.g., tristable, bistable with different potentials) or other fault detection applications beyond bearings would be a logical next step.
  • Investigating the method's robustness against different types of noise and signal interference could enhance its practical utility.

Summary updated

Abstract

Although stochastic resonance (SR) has been widely used to enhance weak fault signatures in machinery and has obtained remarkable achievements in engineering application, the parameter optimization of the existing SR-based methods requires the quantification indicators dependent on prior knowledge of the defects to be detected; for example, the widely used signal-to-noise ratio easily results in a false SR and decreases the detection performance of SR further. These indicators dependent on prior knowledge would not be suitable for real-world fault diagnosis of machinery where their structure parameters are unknown or are not able to be obtained. Therefore, it is necessary for us to design a type of SR method with parameter estimation, and such a method can estimate these parameters of SR adaptively by virtue of the signals to be processed or detected in place of the prior knowledge of the machinery. In this method, the triggered SR condition in second-order nonlinear systems and the synergic relationship among weak periodic signals, background noise and nonlinear systems can be considered to decide parameter estimation for enhancing unknown weak fault characteristics of machinery. Bearing fault experiments were performed to demonstrate the feasibility of the proposed method. The experimental results indicate that the proposed method is able to enhance weak fault characteristics and diagnose weak compound faults of bearings at an early stage without prior knowledge and any quantification indicators, and it presents the same detection performance as the SR methods based on prior knowledge. Furthermore, the proposed method is more simple and less time-consuming than other SR methods based on prior knowledge where a large number of parameters need to be optimized. Moreover, the proposed method is superior to the fast kurtogram method for early fault detection of bearings.
BibTeX
@inproceedings{xu2023stochastic,
  author = {Min Xu and Chao Zheng and Kelei Sun and Li Xu and Zijian Qiao and Zhihui Lai},
  title = {{Stochastic Resonance with Parameter Estimation for Enhancing Unknown Compound Fault Detection of Bearings}},
  booktitle = {Sensors},
  year = {2023},
  url = {https://openreview.net/forum?id=HBP6rhbuBB},
}
Figure