Fault Diagnosis for Rotating Machinery: A Method based on Image Processing.

Area of Science:

• Mechanical Engineering
• Signal Processing
• Machine Learning

Background:

• Rotating machinery is crucial in industrial applications, necessitating effective condition monitoring and fault diagnosis.
• Traditional fault diagnosis methods often rely on 1D signal analysis, which can be limited.
• Multidisciplinary approaches are emerging as a focus for advanced fault diagnosis.

Purpose of the Study:

• To develop and validate a novel multi-discipline, image-processing-based method for the fault diagnosis of rotating machinery.
• To enable automatic feature extraction and diagnosis in a 2D space, overcoming limitations of 1D analysis.
• To demonstrate the method's effectiveness on practical rotating machinery examples.

Main Methods:

• Vibration signals are converted into bi-spectrum contour maps using bi-spectrum technology.
• Speeded-up robust features (SURF) are employed for automatic feature extraction from the 2D maps.
• t-Distributed Stochastic Neighbor Embedding (t-SNE) is used for feature vector dimensionality reduction.
• Probabilistic neural networks (PNN) are utilized for final fault identification.

Main Results:

• The proposed image-processing method successfully extracts fault features from bi-spectrum contour maps.
• Dimensionality reduction using t-SNE effectively highlights key fault features.
• The probabilistic neural network achieves high accuracy in fault identification.
• The method demonstrated high diagnostic accuracy on axial piston hydraulic pumps and self-priming centrifugal pumps.

Conclusions:

• The developed image-processing method offers a highly effective approach for rotating machinery fault diagnosis.
• This multi-discipline technique provides a robust alternative to traditional 1D analysis methods.
• The combination of bi-spectrum analysis, SURF, t-SNE, and PNN shows significant promise for industrial applications.