Controlled Symmetry with Woods-Saxon Stochastic Resonance Enabled Weak Fault Detection

Jian Liu, Jiaqi Guo, Bing Hu, Qiqing Zhai, Can Tang, Wanjia Zhang2023 · DOI

TL;DR

This paper introduces a novel "Controlled Symmetry with Woods-Saxon Stochastic Resonance" (CSwWSSR) model for detecting weak faults in mechanical systems, even under significant noise. By controlling the shape of the potential function with multiple parameters and optimizing them using Particle Swarm Optimization (PSO), the CSwWSSR model significantly enhances the signal-to-noise ratio (SNR) for fault detection compared to existing methods.

What Problem Does This Paper Address?

Detecting subtle faults in mechanical equipment is crucial for ensuring safety and reliability in industrial settings. However, these weak fault signals are often buried under substantial background noise, making them incredibly difficult to identify using traditional signal processing techniques. Existing methods like filtering or decomposition often suppress the noise but can also inadvertently diminish the weak fault signals themselves. Stochastic Resonance (SR) offers a promising alternative by leveraging noise to amplify weak signals, but conventional SR models, like the bistable SR (CBSR), have limitations in their potential function's structure and parameter control, hindering their effectiveness in complex, noisy environments. This paper aims to overcome these limitations by proposing a more flexible and controllable SR model.

Methods

The core of this paper is the development of the Controlled Symmetry with Woods-Saxon Stochastic Resonance (CSwWSSR) model. This model builds upon the Woods-Saxon Stochastic Resonance (WSSR) potential, which allows for individual control over the depth, width, and steepness of its potential wells. The CSwWSSR further enhances this by incorporating a controlled symmetric stochastic resonance (CSSR) component, which introduces two additional potential wells. Crucially, the CSwWSSR model allows for independent control of these three potential wells through distinct parameters: (H, W, a) for the WSSR component and 'k' for the CSSR component.

To find the optimal combination of these parameters for a given signal and noise environment, the authors employ Particle Swarm Optimization (PSO). PSO is used to maximize the output Signal-to-Noise Ratio (SNR), which serves as the objective function for fault detection performance. The process involves:

  1. Preprocessing: Scaling down high-frequency signals to meet the adiabatic approximation theory requirements for SR.
  2. PSO Optimization: Iteratively adjusting the CSwWSSR parameters (H, W, a, k) to maximize the output SNR.
  3. CSwWSSR Calculation: Using the optimized parameters to process the signal via a fourth-order Runge-Kutta method to solve the Langevin equation.
  4. Fault Diagnosis: Analyzing the power spectrum of the output signal to detect fault characteristic frequencies.

Datasets

The paper evaluates the proposed CSwWSSR model using two types of datasets:

  • Simulated Signals: A synthetic signal is generated with a specific fault characteristic frequency (140 Hz) and additive Gaussian white noise. The noise intensity and fault frequency are varied to test the model's robustness.
  • Experimental Bearing Data: Real-world data from a rolling bearing is used, provided by Case Western Reserve University. This dataset includes measurements from a bearing with an outer race fault, sampled at 12 kHz with a rotation speed of 1796 r/min. Additive noise is introduced to simulate a challenging detection scenario.

Key Insights and Ideas

  • Decoupled Parameter Control: The CSwWSSR model's key innovation is its ability to control different aspects of the potential structure independently. Parameters (H, W, a) influence the central well (like WSSR), while 'k' controls the outer wells (CSSR). This granular control is vital for tailoring the SR response to specific signal and noise characteristics.
  • Synergy of WSSR and CSSR: The CSwWSSR model effectively combines the strengths of WSSR (stable particle motion) and CSSR (controlled adjustment of potential wells), leading to superior performance in both noise suppression and signal enhancement.
  • PSO for Optimal Tuning: The use of PSO is critical for finding the optimal parameter set for the CSwWSSR model, allowing it to adapt to diverse fault signatures and noise levels. This automated optimization is a significant advantage over manual tuning.
  • Visualizing Potential Functions: The paper provides clear visualizations of how each parameter (H, W, a, k) affects the shape of the potential function, offering valuable intuition into the model's behavior.

Key Results

  • Improved SNR: In simulation experiments, the CSwWSSR model (both depth and width controlled variants) achieved significantly higher output SNRs compared to standalone WSSR and CSSR models. The CSwWSSR-width model achieved an SNR of -9.777 dB, outperforming WSSR (-10.16 dB) and CSSR variants (-10.85 dB and -11.63 dB).
  • Enhanced Fault Characteristic Amplitude: The CSwWSSR model consistently produced higher amplitudes for the fault characteristic frequency in the output spectrum for both simulated and real bearing data. For the bearing data, CSwWSSR-width achieved an amplitude of 0.6009, surpassing WSSR (0.5423) and CSSR variants (0.4721 and 0.4431).
  • Robustness to Noise and Frequency: The CSwWSSR model demonstrated superior performance across a range of noise intensities and fault characteristic frequencies in simulation tests, consistently yielding higher SNRs than WSSR and CSSR.
  • Practical Applicability: The experimental demonstration on bearing data confirmed the CSwWSSR's effectiveness in a real-world scenario, showing its capability to detect outer race faults even when noise is intentionally added.

Limitations and Future Work

  • Adiabatic Approximation: The paper acknowledges that for high-frequency signals, the adiabatic approximation theory of SR may not hold. A rescaling method is used to address this, but it's a workaround rather than a fundamental solution for all SR scenarios.
  • Future Work: The authors propose extending this work to optimize two-dimensional signals using multistable SR. They highlight the challenge of utilizing SR to overcome spatial correlations in 2D signals and environmental noise in low SNR conditions, suggesting this as a promising avenue for future research.

Summary updated

BibTeX
@misc{liu2023controlled,
  author = {Jian Liu and Jiaqi Guo and Bing Hu and Qiqing Zhai and Can Tang and Wanjia Zhang},
  title = {{Controlled Symmetry with Woods-Saxon Stochastic Resonance Enabled Weak Fault Detection}},
  year = {2023},
  doi = {10.3390/s23115062},
}
Figure