Convenient intelligent diagnosis for rotating machinery: An improved deep forest method based on feature...

Area of Science:

• Engineering
• Computer Science
• Machine Learning

Background:

• Deep learning offers advantages in automated feature extraction for intelligent fault diagnosis.
• Complex hyper-parameter tuning and limited data hinder practical applications, especially for multiple mixed faults in rotating machinery.

Purpose of the Study:

• To develop a convenient and effective intelligent fault diagnosis method for rotating machinery.
• To address challenges of high computational cost, feature submergence, and small training samples in fault diagnosis.

Main Methods:

• An improved deep forest model is proposed for intelligent fault diagnosis.
• A feature reconstruction algorithm is integrated to handle long time-series vibration data and mitigate feature submergence.
• The method is validated against deep neural network-based approaches.

Main Results:

• The proposed improved deep forest method demonstrates superior effectiveness in diagnosing faults.
• The method shows robustness across various hyper-parameter settings.
• Experimental results confirm the superiority over traditional deep learning methods for rotating machinery fault diagnosis.

Conclusions:

• The improved deep forest method offers a practical and robust solution for intelligent fault diagnosis of rotating machinery.
• The feature reconstruction algorithm effectively addresses challenges associated with vibration data and limited samples.
• This approach enhances the applicability of deep learning in industrial diagnostics.