Intelligent Fault Diagnosis of Rolling Element Bearings Based on Modified AlexNet

Area of Science:

• Engineering
• Artificial Intelligence
• Machine Learning

Background:

• Early detection of bearing faults is crucial for industrial system safety and reliability.
• Traditional deep learning convolutional neural network (CNN) models face challenges like excessive parameters, overfitting, and reduced efficiency in noisy or variable load conditions.

Purpose of the Study:

• To develop a novel technique for accurate and efficient bearing fault diagnosis.
• To address the limitations of traditional CNNs, including parameter quantity, overfitting, and performance degradation in real-world industrial environments.

Main Methods:

• Identified the optimal pre-trained CNN model for bearing fault classification.
• Modified the selected CNN model to reduce parameters, mitigate overfitting, and decrease computation time.
• Developed a hybrid model by integrating the best classifier for enhanced performance.

Main Results:

• The proposed technique demonstrates high reliability and efficiency in detecting and classifying bearing faults.
• The method performs effectively under varying load conditions and in noisy environments with different signal-to-noise ratio (SNR) values.

Conclusions:

• The novel hybrid CNN approach offers a robust solution for intelligent bearing fault diagnosis.
• This technique significantly improves upon traditional CNN methods, providing accurate and efficient fault identification in industrial settings.