Bearing Fault Diagnosis with 22-Dimensional Vibration Features

The dataset supports a paper proposing an ISMO-PNN framework for bearing fault diagnosis in robotic systems. It is based on the CWRU bearing dataset, where a 22-dimensional mixed feature set was extracted from vibration signals and reduced to three dimensions while retaining over 80% of discriminative information. The proposed model achieved a fault classification accuracy of 97.14%.

Use Cases

• Train a Probabilistic Neural Network (PNN) using the 22-dimensional mixed feature set extracted from vibration signals for bearing fault classification.
• Apply the Improved Spider Monkey Optimization (ISMO) algorithm to automatically optimize the smoothing parameters of a PNN model for fault diagnosis.
• Use the three-dimensional features, reduced via intelligent PCA, to build a simpler classifier while retaining discriminative information.
• Benchmark fault diagnosis models against the reported 97.14% accuracy and 97.32% macro-average F1 score achieved on this data.

Strengths

• The derived feature set retains over 80% of discriminative information after dimensionality reduction from 22 to 3 dimensions.
• The associated model demonstrates strong generalization with a minimum training-testing accuracy difference of 0.72%.
• The framework achieved a high fault classification accuracy of 97.14% on the CWRU bearing dataset.

Limitations

• The dataset is described within a 676.6 KB DOCX file, indicating it is likely a summary or paper, not the raw or processed tabular data itself.
• Specific row count, column names, and sample data are unavailable, limiting direct analytical use.
• The temporal and geographic coverage of the underlying CWRU data is not specified in the provided description.

Provenance

Source

Fan Zhang, via figshare.

Collection Method

Features were extracted from the CWRU bearing dataset vibration signals, with dimensionality reduction via intelligent PCA.

Freshness

The associated resource was last updated on 2026-03-20.