Fault Diagnosis of Rotary Machines Using Deep Convolutional Neural Network with Wide Three Axis Vibration Signal...

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Signal Processing

Background:

• Early fault detection in rotary machinery is crucial for cost and time savings.
• Traditional fault diagnosis methods can be complex and time-consuming.
• Data-driven approaches offer potential for automated and accurate diagnostics.

Purpose of the Study:

• To develop a novel deep-learning-based fault diagnosis technique for rotary machinery.
• To utilize raw three-axis accelerometer signals as input for a convolutional neural network.
• To achieve high classification accuracy for different machine states.

Main Methods:

• A deep learning model, specifically a convolutional neural network (CNN), was developed.
• Raw three-axis accelerometer data was processed as a high-definition 1D image (6400x1x3 matrix).
• The number of kernels in the CNN was optimized using a grid search approach.

Main Results:

• The proposed technique effectively extracts signal features automatically using deep learning layers.
• The CNN model demonstrated high classification accuracy for identifying various rotary machinery states.
• The use of a wide input matrix (6400x1x3) contributed to good classification performance.

Conclusions:

• The developed deep-learning-based technique provides an effective method for data-driven fault diagnosis in rotary machinery.
• Convolutional neural networks are well-suited for analyzing vibration data from accelerometers for fault classification.
• This approach offers a promising avenue for automated and accurate machinery health monitoring.