High-Performance Adaptive Weak Fault Diagnosis Based on the Global Parameter Optimization Model of a Cascaded Stochastic Resonance System

Zhihui Lai, Zhangjun Huang, Min Xu, Chen Wang, Junchen Xu, Cailiang Zhang, Ronghua Zhu, Zijian QiaoSensors 2023 · OpenReview

TL;DR

This paper introduces a novel "Global Parameter Optimization" (GPO) model for cascaded stochastic resonance (SR) systems to significantly improve the detection of weak fault signals in machinery. By optimizing all parameters of the cascaded system simultaneously, their GPO model outperforms existing methods, demonstrating its effectiveness in real-world bearing fault diagnosis.

What Problem Does This Paper Address?

Detecting subtle faults in machinery, especially in noisy environments, is a critical challenge for ensuring operational safety and preventing catastrophic failures. Traditional signal processing methods often struggle to isolate these weak fault signatures from strong background noise, sometimes even diminishing the fault information itself. Stochastic Resonance (SR) offers a promising alternative by leveraging noise to enhance weak signals. However, the detection capabilities of single SR systems are limited, and existing methods for optimizing cascaded SR systems (multiple SR systems in series) are not fully effective in achieving the theoretically optimal performance. This paper aims to bridge this gap by proposing a more comprehensive optimization strategy for cascaded SR systems to achieve high-performance weak fault diagnosis.

Methods

The core of this work is the development of a Global Parameter Optimization (GPO) model for cascaded stochastic resonance (SR) systems. Unlike previous "fixed-parameter optimization" (FPO) models where parameters are identical across subsystems, the GPO model treats each subsystem's parameters independently. This allows for a more holistic optimization of the entire cascaded system.

The optimization process is driven by the Particle Swarm Optimization (PSO) algorithm, a well-established metaheuristic. The objective function for the PSO is the Signal-to-Noise Ratio (SNR) of the output signal, aiming to maximize it. The paper explores two types of SR systems within the cascade: bistable and tri-stable potential functions, noting that tri-stable systems generally offer better performance. They also investigate the influence of the number of cascaded subsystems, finding that a four-stage system offers a good balance between performance and computational cost.

Datasets

The paper utilizes simulated weak signals for initial testing and then validates the proposed method on two real-world datasets:

  • Rolling Bearing Inner Raceway Fault Data: Obtained from the Case Western Reserve University (CWRU) bearing data center. This dataset involves a simulated fault (a groove cut into the inner raceway) on a deep groove ball bearing.
  • Rolling Bearing Outer Raceway Fault Data: This dataset involves a simulated fault (a groove cut into the outer raceway) on a cylindrical roller bearing.

Key Insights and Ideas

  • Global Optimization is Key: The most significant insight is that optimizing all parameters of a cascaded SR system jointly (GPO) yields superior weak signal detection compared to optimizing parameters within each subsystem independently or using fixed parameters across all subsystems. This is because the cascaded system as a whole has a unique optimal operating point that cannot be reached by optimizing subsystems in isolation.
  • Tri-stable SR Outperforms Bistable: The study consistently shows that SR systems employing tri-stable potential functions offer better weak signal detection performance than those with bistable potential functions. This suggests that the increased complexity of the tri-stable potential landscape is more effective in harnessing noise for signal enhancement.
  • Cascaded Systems are Powerful: The research reinforces the idea that cascading multiple SR systems significantly boosts weak signal detection capabilities compared to single SR systems, especially for signals with very low initial SNRs.
  • Four-Stage System is Optimal: The investigation into the number of subsystems revealed that while increasing the number of stages generally improves performance, it plateaus and can even degrade beyond a certain point (around 4-6 stages). This is attributed to the increased risk of local optima and higher computational demands. A four-stage system strikes a good balance.
  • Robustness to Low SNR: The GPO model demonstrates remarkable robustness and stability in detecting signals with extremely low input SNRs, maintaining a relatively stable output SNR even as the input SNR decreases.

Key Results

  • Improved SNR: The GPO model achieved significantly higher output SNRs compared to single SR systems and FPO models. For instance, in one simulation, the GPO model achieved an SNRout of -0.72 dB, compared to -3.43 dB for a single bistable system and -1.16 dB for the FPO model.
  • Real-World Fault Detection: The proposed GPO model successfully identified characteristic fault frequencies in both simulated and real-world rolling bearing fault datasets, even when these frequencies were heavily submerged in noise (e.g., input SNRs of -39.45 dB and -57.80 dB).
  • For the inner raceway fault, the model detected a frequency of 164.1 Hz (close to the theoretical 160.1 Hz) with an output SNR of +3.31 dB.
  • For the outer raceway fault, the model detected a frequency of 80.57 Hz (close to the theoretical 82.95 Hz) with an output SNR of -11.33 dB.
  • Statistical Significance: Statistical analysis over 100 simulations confirmed the superiority of the GPO model and the tri-stable SR systems, showing higher average, maximum, and minimum output SNRs.

Limitations and Future Work

  • Computational Cost: While the GPO model offers superior performance, it involves optimizing a larger number of parameters, which can increase computational time compared to simpler models. The paper addresses this by identifying the four-stage system as a cost-effective compromise.
  • Parameter Range Determination: The paper mentions the need to determine appropriate search ranges for parameters (like ɛ and R) to satisfy small-parameter conditions. While the study provides ranges, further research could explore more automated or adaptive methods for this.
  • Generalizability: While the paper focuses on rolling bearing faults, future work could explore the application of this GPO model to other types of mechanical faults and machinery.
  • Real-world Noise: The study primarily uses simulated noise. Investigating the performance with complex, non-Gaussian, or time-varying real-world noise could be a valuable extension.

Summary updated

Abstract

Stochastic resonance (SR), as a type of noise-assisted signal processing method, has been widely applied in weak signal detection and mechanical weak fault diagnosis. In order to further improve the weak signal detection performance of SR-based approaches and realize high-performance weak fault diagnosis, a global parameter optimization (GPO) model of a cascaded SR system is proposed in this work. The cascaded SR systems, which involve multiple multi-parameter-adjusting SR systems with both bistable and tri-stable potential functions, are first introduced. The fixed-parameter optimization (FPO) model and the GPO models of the cascaded systems to achieve optimal SR outputs are proposed based on the particle swarm optimization (PSO) algorithm. Simulated results show that the GPO model is capable of achieving a better SR output compared to the FPO model with rather good robustness and stability in detecting low signal-to-noise ratio (SNR) weak signals, and the tri-stable cascaded SR system has a better weak signal detection performance compared to the bistable cascaded SR system. Furthermore, the weak fault diagnosis approach based on the GPO model of the tri-stable cascaded system is proposed, and two rolling bearing weak fault diagnosis experiments are performed, thus verifying the effectiveness of the proposed approach in high-performance adaptive weak fault diagnosis.
BibTeX
@inproceedings{lai2023high,
  author = {Zhihui Lai and Zhangjun Huang and Min Xu and Chen Wang and Junchen Xu and Cailiang Zhang and Ronghua Zhu and Zijian Qiao},
  title = {{High-Performance Adaptive Weak Fault Diagnosis Based on the Global Parameter Optimization Model of a Cascaded Stochastic Resonance System}},
  booktitle = {Sensors},
  year = {2023},
  url = {https://openreview.net/forum?id=hJiQ7sv0OS},
}
Figure