Sensor Data-Driven Bearing Fault Diagnosis Based on Deep Convolutional Neural Networks and S-Transform

Area of Science:

• Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Accurate bearing fault diagnosis is vital for rotating machinery reliability and maintenance.
• Traditional methods using Convolutional Neural Networks (CNNs) struggle with 1D sensor data, requiring complex signal processing and manual intervention.
• Existing approaches often lack feasibility and robustness due to increased manual work and reliance on expertise.

Purpose of the Study:

• To propose a novel, data-driven fault diagnosis method for bearings.
• To overcome the limitations of standard CNNs in processing 1D sensor data.
• To enhance the feasibility and robustness of bearing fault diagnosis systems.

Main Methods:

• Developed a novel ST-CNN model by fusing the S-transform (ST) algorithm with CNN.
• Designed an ST layer to automatically convert 1D sensor data into 2D time-frequency matrices, eliminating manual conversion.
• Trained the ST-CNN model using time-frequency coefficient matrices for fault classification via a softmax layer.

Main Results:

• The proposed ST-CNN method demonstrated superior diagnostic performance compared to existing methods.
• The method achieved higher accuracy and robustness in bearing fault diagnosis.
• Experimental validation was performed using two publicly available bearing datasets.

Conclusions:

• The ST-CNN method offers an effective and automated approach to bearing fault diagnosis.
• Automatic conversion of sensor data to time-frequency matrices enhances diagnostic capabilities.
• The proposed method presents a more feasible and robust solution for rotating machinery maintenance.