Deep learning algorithms for rotating machinery intelligent diagnosis: An open source benchmark study.

Area of Science:

• Engineering
• Computer Science
• Artificial Intelligence

Background:

• Deep learning (DL) shows promise for intelligent diagnosis of rotating machinery, potentially reducing equipment failures.
• Current research lacks standardized datasets, hyper-parameters, and open-source code, hindering fair model comparison and progress.

Purpose of the Study:

• To establish a comprehensive benchmark for evaluating deep learning models in rotating machinery intelligent diagnosis.
• To provide an open-source code library for reproducible and fair model comparisons.
• To identify and address key challenges in the field, including class imbalance, generalization, interpretability, and few-shot learning.

Main Methods:

• Evaluated four deep learning models: Multi-layer Perception (MLP), Auto-Encoder (AE), Convolutional Neural Network (CNN), and Recurrent Neural Network (RNN).
• Utilized nine publicly available datasets with variations in data split strategies, input formats, normalization, and augmentation methods.
• Developed and released an integrated code library for standardized evaluation and comparison.

Main Results:

• Established a benchmark study comparing DL models across multiple datasets and configurations.
• Identified specific challenges impacting model performance, such as class imbalance and generalization.
• Demonstrated the utility of the open-source code library for fair and efficient model testing.

Conclusions:

• The developed benchmark and open-source code facilitate reproducible research and accelerate progress in rotating machinery intelligent diagnosis.
• Addressing issues like class imbalance and generalization is crucial for robust DL model deployment.
• Standardized evaluation frameworks are essential for advancing the field of intelligent diagnosis.