Fault diagnosis of rotor based on Semi-supervised Multi-Graph Joint Embedding

Area of Science:

• Engineering
• Data Science
• Machine Learning

Background:

• Traditional graph embedding methods inadequately represent complex, multi-sample relationships in high-dimensional fault data.
• Hypergraph construction can suffer from an "averaging effect" that distorts similarity relationships, reducing fault classification accuracy.
• Accurate characterization of high-dimensional fault data structures is crucial for effective diagnostic systems.

Purpose of the Study:

• To propose a novel dimensionality reduction method, Semi-supervised Multi-Graph Joint Embedding (SMGJE), for rotor fault diagnosis.
• To address the limitations of traditional hypergraph construction in representing complex fault data relationships.
• To enhance the accuracy of fault classification by overcoming the "averaging effect" in hypergraph embeddings.

Main Methods:

• Developed Semi-supervised Multi-Graph Joint Embedding (SMGJE) for dimensionality reduction.
• Constructed both simple graphs and hypergraphs using the same sample points to capture data structure.
• Employed a multi-graph joint embedding approach to characterize high-dimensional data.
• Utilized direct similarity descriptions in simple graph edges to mitigate hypergraph "averaging effects".

Main Results:

• SMGJE effectively overcomes the "averaging effect" inherent in traditional hypergraph construction.
• The proposed method accurately portrays the complex structural relationships within high-dimensional fault data.
• Validation on two distinct fault datasets demonstrated the effectiveness of SMGJE for rotor fault diagnosis.

Conclusions:

• SMGJE offers a superior approach to dimensionality reduction for high-dimensional fault data compared to traditional methods.
• The multi-graph joint embedding strategy enhances the representation of complex data structures, leading to improved fault classification.
• This method provides a promising advancement for accurate and reliable rotor fault diagnosis systems.