A Two-Stage, Intelligent Bearing-Fault-Diagnosis Method Using Order-Tracking and a One-Dimensional Convolutional Neural Network with Variable Speeds

Mengyu Ji, Gaoliang Peng, Jun He, Shaohui Liu, Zhao Chen, Sijue LiSensors 2021 · OpenReview

TL;DR

This paper introduces a novel two-stage method, OT-1DCNN, for diagnosing bearing faults under varying rotational speeds. By combining order-tracking for signal normalization and a 1D convolutional neural network for feature extraction, the method achieves high accuracy even when trained on data from a single speed, significantly improving upon existing techniques for this challenging problem.

What Problem Does This Paper Address?

Diagnosing faults in rotating machinery like bearings is crucial for preventing catastrophic failures. A major challenge in this field is that the vibration signals produced by a faulty bearing become "blurred" in the frequency domain when the bearing's rotation speed changes. This blurring makes it difficult to extract consistent fault signatures, leading to reduced diagnostic accuracy for traditional and even many intelligent methods. Existing approaches often require extensive training data across various speeds or struggle to generalize to new speed conditions. This paper tackles the problem of robust bearing fault diagnosis under variable speed conditions, aiming to develop a method that is accurate and requires less comprehensive training data.

Methods

The proposed solution, OT-1DCNN, employs a two-stage approach:

  1. Order-Tracking Resampling: This stage addresses the variable speed issue directly. Order tracking is used to resample the raw vibration data. Instead of sampling at fixed time intervals, it samples based on the rotational angle of the bearing. This process effectively converts the non-stationary vibration signal into an angular domain signal, normalizing it to a consistent "order" regardless of the actual rotation speed. This makes the fault characteristic frequencies stable and comparable across different speeds. The paper details a computed order tracking algorithm, using linear interpolation for resampling.
  2. 1D Convolutional Neural Network (1D-CNN) for Feature Extraction and Classification: After resampling, the data is fed into a 1D-CNN. Unlike traditional 2D-CNNs often used for image data, this 1D-CNN is designed to process sequential time-series data like vibration signals. The network is architected with multiple convolutional layers to automatically extract relevant fault features from the resampled data. The paper highlights the use of wide kernels in the initial layer to capture broader spectral information and smaller kernels in subsequent layers for finer details. The extracted features are then passed through fully connected layers for final fault classification. A key aspect is that the network is trained on data collected at one specific speed and then used to classify data from other speeds, demonstrating its generalization capability.

Datasets

The proposed method was evaluated on two distinct datasets:

  • Case Western Reserve University (CWRU) Bearing Dataset: This is a widely used benchmark dataset for bearing fault diagnosis. The experiments utilized data collected under various motor loads and speeds, with different fault types (normal, inner race, outer race, ball). The paper created four specific datasets (A, B, C, D) from this source, where training data was from a single speed (1730 RPM) and testing data was from multiple speeds (1750, 1772, 1797 RPM).
  • Own Experimental Platform Dataset: The authors also built their own self-made bearing fault platform to generate experimental data. This dataset included health status, inner ring fault, and outer ring fault data collected at four different speeds (900, 1020, 1140, 1260 RPM). Similar to the CWRU experiments, training was performed on data from one speed (900 RPM), and testing on data from other speeds. This dataset allowed for validation under conditions with potentially larger speed discrepancies between training and testing.

Key Insights and Ideas

  • Order-Tracking as a Universal Preprocessor: The paper strongly advocates for order tracking as a fundamental preprocessing step for bearing fault diagnosis under variable speeds. It effectively "normalizes" the signal's spectral characteristics to an angular domain, making them invariant to rotational speed changes.
  • 1D-CNN for Efficient Feature Learning: By leveraging 1D-CNNs, the method automates the complex and often manual process of feature engineering, allowing the network to learn discriminative fault features directly from the vibration data.
  • Single-Speed Training for Multi-Speed Diagnosis: The most significant insight is the successful demonstration of training a diagnostic model on data from just one speed and achieving high diagnostic accuracy on data from other speeds. This drastically reduces the burden of collecting comprehensive, multi-speed training datasets.
  • Synergy of Order-Tracking and CNNs: The combination of order tracking's signal normalization and CNN's adaptive feature extraction creates a powerful synergy, addressing both the speed variation problem and the complexity of fault signature identification.

Key Results

The OT-1DCNN method demonstrated superior performance compared to baseline methods, particularly in variable speed conditions:

  • CWRU Dataset: In conditions with the largest speed difference, the OT-1DCNN achieved accuracies higher than baseline DCNN by 0.74% (Dataset A), 2.62% (Dataset B), and 4.03% (Dataset C). When compared against DCNN, AlexNet, and VGGNet, OT-1DCNN consistently achieved the highest accuracy across datasets, with improvements of up to 2.62% over DCNN on Dataset B.
  • Own Dataset: On the self-made platform, the speed discrepancies were larger. OT-1DCNN showed significant improvements over DCNN, with accuracy gains of 3.21% (Dataset A), 7.27% (Dataset B), and 11.00% (Dataset C). The paper highlights that the improvement rate increases with a larger speed difference between training and testing data.
  • Generalization: The method consistently performed well in migration diagnosis, meaning it could accurately classify faults from data collected at speeds different from the training speed.

Limitations and Future Work

  • Speed Discrepancy Impact: The authors acknowledge that while the method is robust, the diagnostic accuracy does decrease to some extent as the speed difference between training and testing data increases.
  • Future Work: The authors plan to further investigate the reasons for this accuracy decline with larger speed discrepancies and explore ways to mitigate it. They also aim to continue exploring improvements in network classification accuracy for future research.

Summary updated

Abstract

When performing fault diagnosis tasks on bearings, the change of any bearing’s rotation speed will cause the frequency spectrum of bearing fault characteristics to be blurred. This makes it difficult to extract stable fault features based on manual or intelligent methods, resulting in a decrease in diagnostic accuracy. In this paper, a two-stage, intelligent fault diagnosis method (order-tracking one-dimensional convolutional neural network, OT-1DCNN) is proposed to deal with the problem of fault diagnosis under variable speed conditions. Firstly, the order tracking algorithm is used to resample the monitoring data obtained under different rotation speeds. Then, the one-dimensional convolutional neural network is adopted to extract features of the fault data. Finally, the fault type of collected data can be obtained by fully connected networks based on the features extracted. In the time domain, while the proposed algorithm only relies on the fault data collected under one speed as the training dataset, it is capable of doing fault diagnosis under different speed conditions. In the condition with the largest difference in speed with each dataset, the accuracy of the proposed method is higher than the baseline methods by 0.54% and 11.00%—on CWRU dataset and our own dataset respectively. The results show that the proposed method performs well in dealing with the fault diagnosis under the condition of variable speeds.
BibTeX
@inproceedings{ji2021two,
  author = {Mengyu Ji and Gaoliang Peng and Jun He and Shaohui Liu and Zhao Chen and Sijue Li},
  title = {{A Two-Stage, Intelligent Bearing-Fault-Diagnosis Method Using Order-Tracking and a One-Dimensional Convolutional Neural Network with Variable Speeds}},
  booktitle = {Sensors},
  year = {2021},
  url = {https://openreview.net/forum?id=K5eRoFT5Csge},
}
Figure