Single-Sensor Engine Multi-Type Fault Detection.

Area of Science:

• Engineering
• Data Science
• Signal Processing

Background:

• Engine fault detection is crucial for equipment reliability and cost reduction.
• Obtaining high-quality data is challenging, often limiting analysis to single-sensor inputs.
• Existing methods may struggle with data scarcity and diverse operating conditions.

Purpose of the Study:

• To develop a robust engine fault detection method using limited single-sensor data.
• To enhance the effectiveness of Variational Mode Decomposition (VMD) for feature extraction.
• To improve classification accuracy and training efficiency using Random Forest (RF).

Main Methods:

• An improved VMD (IVMD) was developed, optimizing mode number and penalty terms for better signal decomposition.
• Feature vectors were engineered using attributes like energy, frequency, and singular values.
• Feature selection based on importance ranking was applied to Random Forest (RF) classification.

Main Results:

• The IVMD effectively decomposed engine signals, addressing mode aliasing and improving efficiency.
• The proposed VMD-RF method achieved higher accuracy in fault diagnosis across various conditions (single-speed, multi-speed, cross-speed).
• The method demonstrated faster training efficiency compared to other algorithms, especially with limited data.

Conclusions:

• The combined IVMD and RF approach offers a promising solution for engine fault detection with minimal data and hardware.
• This method significantly improves diagnostic accuracy and efficiency, applicable to real-world scenarios.
• The technique holds potential for reducing maintenance costs and enhancing equipment reliability.