A Rotating Machinery Fault Diagnosis Method Based on Feature Learning of Thermal Images

Zhen Jia, Zhenbao Liu, Chi-Man Vong, Michael G. PechtIEEE Access 2019 · OpenReview

Abstract

Rotating machinery is fundamental to industrial operations, where mechanical failures can lead to significant downtime and safety risks. Traditional diagnostic methods primarily rely on vibration signals, which often suffer from sensor installation difficulties and environmental noise. This paper presents a non-contact fault diagnosis approach using infrared thermography (IRT). By leveraging feature learning techniques—specifically Bag-of-Visual-Words (BoVW) and Convolutional Neural Networks (CNN)—the proposed method automatically extracts diagnostic features from thermal images. Experimental results on bearing datasets demonstrate that this IRT-based approach achieves up to 100% accuracy, surpassing traditional vibration-based methods.

Introduction

In industrial maintenance, monitoring the health of rotating machinery like motors and generators is critical. While vibration analysis is the current standard, it presents several challenges:

  1. Intrusive Installation: Sensors must be physically attached, which is often impractical or risky for sensitive equipment.
  2. Signal Interference: Factory floors are noisy environments where electromagnetic interference can degrade vibration data.
  3. Manual Feature Engineering: Previous thermal imaging attempts required manual identification of "hot spots," which often failed to capture the full complexity of mechanical faults.

This research introduces a "hands-off" system that utilizes infrared thermal signatures and deep learning to automate fault detection without physical contact.

Methodology

The researchers developed a comprehensive pipeline to transform raw thermal data into automated diagnoses.

Data Acquisition and Preprocessing

Using a WS-ZHT1 multifunctional rotor test rig, the team captured thermal signatures under nine distinct conditions, including normal operation, unbalance, roller faults, and various compound faults. A total of 900 images were collected. To enhance efficiency, a "Region of Interest" (ROI) approach was used to automatically crop images, focusing specifically on the bearing area.

Feature Learning Strategies

The study evaluated two primary computer vision strategies for feature extraction:

  1. Bag-of-Visual-Words (BoVW): This method identifies "interest points" using SIFT, clusters them into a "visual vocabulary," and generates a histogram of these "words" for classification.
  2. Convolutional Neural Network (CNN): A 5-layer deep learning model designed to automatically learn hierarchical features directly from raw pixels.
Flowchart of the proposed fault diagnosis framework using BoVW and CNN.

The CNN architecture, specifically, was designed to capture spatial relationships of heat across the entire component, moving beyond the local point analysis of traditional methods.

Architecture of the 5-layer Convolutional Neural Network used for feature extraction.

Results and Discussion

The experimental results highlight the superiority of deep learning in thermal fault diagnosis.

  • Diagnostic Accuracy: The CNN achieved a perfect 100% accuracy across all fault states in both transient (initial 5 minutes) and steady-state (after 15 minutes) conditions. The BoVW method also performed strongly, reaching 99.8% accuracy in the steady state.
  • Steady State vs. Transient: Accuracy was generally higher during steady-state operation. In the transient phase, heat has not fully dissipated, making the visual signatures of different faults more similar and harder to distinguish.
  • Comparison with Vibration Methods: In direct comparisons for specific faults (Normal, Roller, Outer Race), the IRT-based CNN consistently outperformed traditional vibration-based sparse coding methods, which typically peaked at 98% accuracy.

Conclusion

This study demonstrates that infrared thermography, when combined with modern feature learning, offers a robust, non-contact alternative to traditional vibration monitoring. By eliminating the need for physical sensors and manual feature design, this method reduces noise interference and simplifies the monitoring of complex industrial systems. Future work will focus on testing these models in more chaotic real-world environments and optimizing computational costs for real-time deployment on edge devices.

Summary updated

Abstract

The rotating machinery plays a vital role in industrial systems, in which unexpected mechanical faults during operation can lead to severe consequences. For fault prevention, many fault diagnostic methods based on vibration signals are available in the literature. However, the vibration signals are obtained by using different types of sensors, which can cause sensor installation issues and damage the rotating machinery. In addition, this kind of data acquisition through vibration signal induces a large amount of signal noise during machine operation, which will challenge the later fault diagnosis. A recent fault detection method based on infrared thermography (IRT) for rotating machinery avoids these issues. However, the corresponding literature is limited by the fact that the characteristics of the manual design cannot characterize the fault completely so that the diagnostic accuracy cannot exceed the diagnostic method based on the vibration signals. This paper introduces a popular image feature extraction method into the fault diagnosis of rotating machinery based on IRT for the first time. First, capturing the IRT images of the rotating machinery in different states, and then two popular feature extraction methods for IRT images, bag-of-visual-word, and convolutional neural network, are tested in turn. Finally, the extracted features are classified to implement the automatic fault diagnosis. The developed method is applied to analyze the experimental IRT images collected from bearings, and the results demonstrate that the developed method is more effective than the traditional methods based on vibration signals.
BibTeX
@inproceedings{jia2019rotating,
  author = {Zhen Jia and Zhenbao Liu and Chi-Man Vong and Michael G. Pecht},
  title = {{A Rotating Machinery Fault Diagnosis Method Based on Feature Learning of Thermal Images}},
  booktitle = {IEEE Access},
  year = {2019},
  url = {https://openreview.net/forum?id=rEBhKVjJuq},
}
Figure