A plug-and-play data processing module for complex faults diagnosis.

Area of Science:

• Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Intelligent fault diagnosis is crucial for modern equipment but limited by small datasets and complex faults.
• Existing methods struggle with data scarcity and intricate fault patterns, hindering widespread industrial application.

Purpose of the Study:

• To develop a plug-and-play data processing module to overcome limitations in intelligent fault diagnosis.
• To enhance deep learning model performance by expanding sample size and improving data relationships.

Main Methods:

• An Amplitude-phase composite data enhancement (APCUP) method using Discrete Fourier Transform to fuse fault types and generate new samples.
• An adjacent label smoothing (ALS) strategy to process labels, preventing over-classification and enhancing model robustness.
• Integration of these methods into a novel data processing module.

Main Results:

• Demonstrated efficacy and viability on an industrial robot arm platform and two public datasets.
• Markedly enhanced model capacity for diagnosing intricate faults.
• Significant increase in diagnostic accuracy and reduction in errors after module incorporation.

Conclusions:

• The developed data processing module effectively expands training datasets and improves the internal data relationships.
• The APCUP and ALS strategies enhance deep learning models for intelligent fault diagnosis, particularly for complex industrial equipment.
• The module offers a practical solution for improving the accuracy and robustness of fault diagnosis systems.