Numerical Model Driving Multi-Domain Information Transfer Method for Bearing Fault Diagnosis

Long Zhang, Hao Zhang, Qian Xiao, Lijuan Zhao, Yanqing Hu, Haoyang Liu, Yu QiaoSensors 2022 · DOI

TL;DR

This paper introduces a novel method for diagnosing rolling bearing faults under varying operating conditions, even with limited real-world fault data. By combining a high-fidelity dynamic simulation model with a cross-domain adaptation technique (cDNAP), they achieve up to 99% diagnostic accuracy, outperforming existing domain adaptation methods.

What Problem Does This Paper Address?

Rolling bearings are critical components in rotating machinery, and their failure can lead to significant damage and safety hazards. Diagnosing these faults is crucial, but real-world challenges make it difficult. Specifically, bearing fault diagnosis suffers from two major issues:

  1. Varying Working Conditions: Bearings rarely operate under constant speed and load. Their performance and the resulting vibration signals change significantly with these variations, making it hard for diagnostic models trained on one set of conditions to generalize to others.
  2. Scarcity of Fault Samples: Obtaining sufficient labeled fault data, especially for rare or severe fault types, is expensive and time-consuming. This data scarcity leads to poor model performance and overfitting.

Existing methods often struggle to address both these challenges simultaneously. While some methods handle variable conditions or limited data, few effectively combine both.

Methods

The proposed solution, a Numerical Model Driving Multi-Domain Information Transfer Method for Bearing Fault Diagnosis, leverages a combination of techniques:

  • High-Fidelity Bearing Dynamic Model: The authors first construct a detailed 2D dynamic model of a NU205 bearing. This model simulates vibration signals for various fault types (normal, outer ring, inner ring, rolling element) under different working conditions (speed and load). This addresses the scarcity of real fault data by generating synthetic, yet realistic, fault samples.
  • Generative Adversarial Network (WGAN-GP): To further augment the limited simulated data and ensure sufficient samples for training, a Wasserstein Generative Adversarial Network with Gradient Penalty (WGAN-GP) is employed. This network expands the simulated datasets, generating synthetic samples that mimic the distribution of the original simulated data.
  • Cross-Domain Nuisance Attribute Projection (cDNAP): This is the core novel contribution. cDNAP is designed to eliminate the discrepancies between simulated (source domain) and measured (target domain) data, specifically addressing the interference from variable working conditions. It achieves this by:
  • Constructing a cross-domain simulation projection matrix.
  • This matrix is learned to project features from both simulated and measured signals into a common space.
  • The projection aims to remove "nuisance attributes" related to working conditions while preserving fault-specific information. This effectively aligns the feature distributions of different domains and conditions.

Datasets

  • Simulated Datasets: Generated using a high-fidelity dynamic model of a NU205 bearing, simulating normal and various fault conditions under multiple working conditions (speed and load combinations). These datasets were further expanded using WGAN-GP.
  • Measured Datasets: Acquired from a homemade rotating machinery simulation test bench, replicating bearing fault scenarios. The paper states that the training datasets (source domain) are simulated samples, and the test datasets (target domain) are measured samples, with no overlap.

Key Insights and Ideas

  • Simulation as a Powerful Data Augmentation Tool: The paper demonstrates that a well-constructed dynamic simulation model can effectively generate realistic fault data, overcoming the practical limitations of collecting real-world samples.
  • cDNAP's Effectiveness in Aligning Domains and Conditions: The key innovation, cDNAP, successfully bridges the gap between simulated and real data and accounts for variations in operating conditions. This is crucial because traditional domain adaptation often struggles with the combined challenge.
  • Nuisance Attribute Projection for Robustness: By focusing on projecting out "nuisance attributes" (i.e., variations due to working conditions), the method ensures that the learned features are more robust and generalizable across different scenarios.
  • The Importance of Feature Alignment: The results highlight that simply using domain adaptation isn't enough; aligning features while preserving fault information is critical for high diagnostic accuracy under complex conditions.

Key Results

  • The proposed cDNAP method achieved a diagnostic accuracy of up to 99% in cross-domain fault diagnosis under variable working conditions.
  • This performance significantly outperformed established domain adversarial neural networks like DANN, DSAN, and DAAN, which achieved lower average accuracies (e.g., DAAN: 82.3%, DANN: 78.4%, DSAN: 70.5% in one comparison).
  • Even NAP (Nuisance Attribute Projection without the cross-domain aspect) showed improvement over traditional methods but was outperformed by cDNAP, indicating the necessity of the cross-domain projection.
  • The method demonstrated effectiveness in both single and compound working condition scenarios.

Limitations and Future Work

  • Model Specificity: The dynamic model is specific to the NU205 bearing. The authors acknowledge that different bearing types and operating conditions would require constructing corresponding dynamic models.
  • Manual Feature Extraction: The current approach relies on manually extracted features (time-domain, wavelet packet decomposition, etc.). This can be a bottleneck and may influence fault diagnosis accuracy.
  • Future Work:
    • Developing more effective and high-fidelity dynamic models for a wider range of bearings.
  • Integrating deep learning for automatic feature extraction, combined with the redundant attribute projection method, to eliminate manual feature engineering and further enhance fault diagnosis accuracy.

Summary updated

Abstract

Given the complexity of the application scenarios of rolling bearing and the severe scarcity of fault samples, a solution to the issue of fault diagnosis under varying working conditions along with the absence of fault samples is required. A numerical model-driven cross-domain fault diagnosis method targeting variable working conditions is proposed based on the cross-Domain Nuisance Attribute Projection (cDNAP). Firstly, the simulation datasets consisting of multiple fault types under variable working conditions are constructed to solve the problem of incomplete fault samples. Secondly, the simulation datasets are expanded by means of generating adversarial network to ensure sufficient samples for subsequent model training. Finally, cDNAP is used to obtain the cross-domain simulation projection matrix, which eliminates the variance in the distribution of measured and simulated sample features under varying working conditions. The experimental results of cross-domain for variable working conditions show that the diagnostic accuracy reaches up to 99%. Compared with DANN, DSAN, and DAAN domain adversarial neural networks, the proposed method performs better in bearing fault diagnosis.
BibTeX
@inproceedings{zhang2022numerical,
  author = {Long Zhang and Hao Zhang and Qian Xiao and Lijuan Zhao and Yanqing Hu and Haoyang Liu and Yu Qiao},
  title = {{Numerical Model Driving Multi-Domain Information Transfer Method for Bearing Fault Diagnosis}},
  booktitle = {Sensors},
  year = {2022},
  doi = {10.3390/s22249759},
}
Figure