Lite and Efficient Deep Learning Model for Bearing Fault Diagnosis Using the CWRU Dataset.

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Bearing defects in rotating machinery cause significant downtime and costs.
• Deep learning models offer effective bearing defect diagnosis but are computationally expensive.
• Existing model optimization methods often reduce classification performance.

Purpose of the Study:

• To propose a new approach for bearing defect diagnosis that reduces input data dimensionality and optimizes model structure simultaneously.
• To develop a lightweight convolutional neural network (CNN) for efficient and accurate bearing defect classification.
• To demonstrate the effectiveness of the proposed method on the CWRU dataset.

Main Methods:

• Downsampling vibration sensor signals to reduce input data dimensionality.
• Constructing spectrograms from downsampled signals for feature extraction.
• Developing a lite CNN model with fixed feature map dimensions for classification.

Main Results:

• The proposed method achieved high classification accuracy with significantly lower input data dimensions compared to existing deep learning models.
• The lite CNN model demonstrated high computational efficiency while maintaining outstanding classification performance.
• The approach outperformed a state-of-the-art model for bearing defect diagnosis under various conditions.

Conclusions:

• The proposed method offers an efficient and accurate solution for bearing defect diagnosis, overcoming the computational challenges of complex deep learning models.
• This approach has the potential for broader applications in analyzing high-dimensional time series data beyond bearing failure diagnosis.