Fault Diagnosis of Mechanical Rolling Bearings Using a Convolutional Neural Network–Gated Recurrent Unit Method with Envelope Analysis and Adaptive Mean Filtering

Huiyi Zhu, Zhen Sui, Jianliang Xu, Yeshen LanProcesses 2024 · DOI

TL;DR

This paper introduces a novel deep learning approach for diagnosing faults in mechanical rolling bearings, even in noisy industrial environments. By combining advanced signal processing (envelope analysis and adaptive mean filtering) with a hybrid Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU) model, they achieve a remarkable 99.25% accuracy, outperforming traditional methods.

What Problem Does This Paper Address?

Mechanical bearings are critical components in countless industrial systems, and their reliable operation is paramount for efficiency and safety. However, these bearings often operate under harsh conditions, leading to wear and potential failures. Detecting these faults early is crucial to prevent cascading issues, equipment downtime, and significant economic losses. A major hurdle in fault diagnosis is the presence of substantial noise in vibration signals collected from industrial environments, which can obscure subtle fault signatures. Traditional manual feature extraction methods are time-consuming and prone to human error.

Methods

The proposed method tackles the noise problem in two stages:

  1. Signal Preprocessing:
    • Envelope Analysis: This technique is used to extract the "envelope" of the vibration signal, which helps to reveal fault information that might be hidden within the noise.
  2. Adaptive Mean Filtering: This filter is applied to further smooth the signal and suppress random noise. It adaptively adjusts its filtering based on local signal characteristics, aiming to preserve important fault features while reducing noise.
  3. Deep Learning Model:
    • CNN-GRU Hybrid: A deep learning architecture is employed, combining the strengths of Convolutional Neural Networks (CNNs) and Gated Recurrent Units (GRUs).
  4. CNN: Acts as a spatial feature extractor, identifying local patterns and subtle details within the preprocessed vibration signals.
  5. GRU: Processes the features extracted by the CNN to capture temporal dependencies and sequential patterns, which are crucial for understanding the dynamic behavior of bearing faults.

This hybrid approach allows the model to learn both spatial and temporal characteristics of the bearing signals for more robust fault diagnosis.

Datasets

The study utilizes the CWRU Bearing Data Center dataset. This dataset is a well-established benchmark for bearing fault diagnosis research and includes vibration data collected from bearings with various fault types (ball, inner race, outer race) and sizes, under no-load conditions. The paper categorizes the data into 10 classes: one normal class and nine fault classes (three types of faults, each with three different sizes).

Key Insights and Ideas

  • Synergy of Signal Processing and Deep Learning: The paper highlights that combining traditional signal processing techniques (envelope analysis, adaptive filtering) with modern deep learning models (CNN-GRU) is highly effective. The preprocessing step significantly enhances the quality of the input data for the deep learning model.
  • CNN for Spatial, GRU for Temporal: The distinct roles of CNNs (spatial feature extraction) and GRUs (temporal dependency modeling) are crucial. This combination allows the model to capture a comprehensive understanding of the fault signatures.
  • Noise Resilience: The adaptive mean filtering, in particular, seems to be a key factor in making the system robust to noisy industrial environments, a common challenge in real-world applications.
  • Visualization as Validation: The use of confusion matrices and t-SNE plots provides strong visual evidence of the model's effectiveness in distinguishing between different fault categories and its ability to learn discriminative features.

Key Results

  • High Accuracy: The proposed CNN-GRU method achieved an impressive average accuracy of 99.25% on the CWRU dataset.
  • Superior Performance: Compared to four other classical fault diagnosis models (RNN, GRU, CNN, and ResNet18), the CNN-GRU model demonstrated superior diagnostic accuracy and generalization ability.
  • Reduced Misclassifications: Confusion matrices showed that the CNN-GRU model had the lowest misclassification rates and the highest correct diagnosis rates across all categories.
  • Effective Feature Extraction: t-SNE visualizations clearly showed that the model progressively refined feature representations across its layers, leading to almost complete separation of data points from different fault types at the output layer.

Limitations and Future Work

  • Limitations:
    • The study was primarily conducted under no-load conditions. While this simplifies analysis by isolating bearing defects, real-world industrial scenarios often involve varying loads.
  • The paper focuses on specific fault types and sizes within the CWRU dataset. Its performance on other types of faults or under different operating conditions (e.g., varying speeds, different lubrication states) is not explicitly detailed.
  • Future Work:
    • Broader Applicability: The authors plan to explore the model's applicability to other types of machinery and under different operational conditions.
  • Real-time and Edge Deployment: Investigating the integration of the model with real-time fault diagnosis systems and edge computing frameworks is a key future direction to enhance its practical value.
  • Scalability: Further research into ensuring robustness and scalability for diverse industrial applications is planned.

Summary updated

Abstract

Rolling bearings are vital components in rotating machinery, and their reliable operation is crucial for maintaining the stability and efficiency of mechanical systems. However, fault detection in rolling bearings is often hindered by noise interference in complex industrial environments. To overcome this challenge, this paper presents a novel fault diagnosis method for rolling bearings, combining Convolutional Neural Networks (CNNs) and Gated Recurrent Units (GRUs), integrated with the envelope analysis and adaptive mean filtering techniques. Initially, envelope analysis and adaptive mean filtering are applied to suppress random noise in the bearing signals, thereby enhancing the visibility of fault features. Subsequently, a deep learning model that combines a CNN and a GRU is developed: the CNN extracts spatial features, while the GRU captures the temporal dependencies between these features. The integration of the CNN and GRU significantly improves the accuracy and robustness of fault diagnosis. The proposed method is validated using the CWRU dataset, with the experimental results achieving an average accuracy of 99.25%. Additionally, the method is compared to four classical fault diagnosis models, demonstrating superior performance in terms of both diagnostic accuracy and generalization ability. The results, supported by various visualization techniques, show that the proposed approach effectively addresses the challenges of fault detection in rolling bearings under complex industrial conditions.
BibTeX
@inproceedings{zhu2024fault,
  author = {Huiyi Zhu and Zhen Sui and Jianliang Xu and Yeshen Lan},
  title = {{Fault Diagnosis of Mechanical Rolling Bearings Using a Convolutional Neural Network–Gated Recurrent Unit Method with Envelope Analysis and Adaptive Mean Filtering}},
  booktitle = {Processes},
  year = {2024},
  doi = {10.3390/pr12122845},
}
Figure