Related Experiment Video
Updated: Jun 24, 2026
08:27
Image Recognition and Parameter Analysis of Concrete Vibration State Based on Support Vector Machine
Published on: January 5, 2024
RGB-based visual encoding of vibration data for gearbox fault diagnosis using U-Net segmentation model
İrfan Kiliç1, Gülşah Karaduman2, Beyda Tasar3
1Department of Software Engineering, Engineering Faculty, Firat University, Elazig, Turkey.
Plos One
|June 22, 2026
Summary
This study converts gearbox vibration data into images for deep learning analysis, achieving 99.87% accuracy in fault diagnosis. This novel image-based approach offers superior performance for signal-based engineering problems.
Area of Science:
- Engineering
- Artificial Intelligence
- Data Science
Background:
- Gearbox gear faults pose significant challenges in industrial machinery maintenance.
- Traditional methods for diagnosing these faults often rely on complex signal processing and machine learning techniques.
- Analyzing numerical vibration data presents limitations in extracting comprehensive fault information.
Purpose of the Study:
- To develop an innovative image-based deep learning approach for diagnosing gearbox gear faults.
- To transform numerical vibration sensor data into a visual format for enhanced analysis.
- To evaluate the effectiveness of deep learning models in classifying gearbox faults using image representations of sensor data.
Main Methods:
- Utilized the Gearbox Fault Diagnosis Dataset from Kaggle, collecting vibration signals from four sensors.
- Processed signals by calculating maximum, minimum, and mean values, then normalizing them to a [0-255] range.
- Mapped normalized values to RGB color channels to generate 256x256 pixel images for each fault category.
- Trained a pre-trained U-Net deep learning model for segmentation using these image representations over 10 epochs.
Main Results:
- Achieved a high classification accuracy of 99.87% for gearbox fault diagnosis.
- Obtained a mean average precision (mAP) score of 99.74%, indicating robust model performance.
- Demonstrated significant advantages of the image-based deep learning approach over traditional machine learning and text-based deep learning methods.
Conclusions:
- Converting numerical sensor data into visual RGB images enables highly accurate fault classification using deep learning.
- The proposed method establishes a new paradigm for solving signal-based engineering problems, particularly in gearbox fault detection.
- This study highlights the potential of image-based deep learning for analyzing non-visual numerical data in engineering applications.
