A New Deep Learning Framework for Imbalance Detection of a Rotating Shaft

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Rotor unbalance is a primary cause of industrial machine vibration, leading to reduced efficiency, component damage, and potential failure.
• Existing classification algorithms for rotor imbalance lack flexibility for varying numbers of classes.
• A need exists for a robust multiclass prediction algorithm for real-time rotor imbalance classification.

Purpose of the Study:

• To develop a novel deep learning (DL) algorithm for detecting rotating shaft unbalance.
• To enable both binary (balanced vs. unbalanced) and multiclass (level of unbalance) identification.
• To create a robust, real-time fault detection mechanism for industrial rotating machinery.

Main Methods:

• Developed a DL algorithm integrating ResNet and Convolutional Neural Network (CNN) for feature extraction.
• Utilized accelerometer vibration sensor data, preprocessed using Fast Fourier Transform (FFT) and Short-Time Fourier Transform (STFT).
• Employed STFT for its feature-rich characteristics, enhancing model performance.

Main Results:

• The developed DL model achieved 99.23% testing accuracy for binary classification (balanced vs. unbalanced).
• The model demonstrated 95.15% testing accuracy for multiclass classification of unbalance levels.
• The proposed algorithm outperformed standalone ResNet and CNN models.

Conclusions:

• The novel DL framework is robust for both binary and multiclass rotor unbalance classification.
• The algorithm offers a reliable solution for real-time fault detection in industrial rotating machinery.
• This approach enhances operational safety and extends the lifespan of critical machine components.