A Novel Fault Diagnosis Method for Rotating Machinery Based on a Convolutional Neural Network

Area of Science:

• Mechanical Engineering
• Signal Processing
• Artificial Intelligence

Background:

• Rotating machinery requires robust fault diagnosis for safety and operational reliability.
• Traditional fault diagnosis methods often reduce signal dimensionality, potentially losing critical information.
• Existing techniques may not fully capture the complex dynamics of fault signatures in vibration data.

Purpose of the Study:

• To propose a novel fault diagnosis method for rotating machinery.
• To leverage convolutional neural networks (CNNs) for direct analysis of time-frequency representations.
• To improve the accuracy and completeness of fault detection in vibration signals.

Main Methods:

• Generated continuous wavelet transform scalograms (CWTS) from raw vibration signals.
• Employed wavelet transform to decompose signals across various scales.
• Trained a convolutional neural network (CNN) to classify CWTS for fault identification.

Main Results:

• The proposed CNN-based method accurately diagnosed faults in experimental rotating machinery.
• Experiments on a rotor platform demonstrated high diagnostic precision.
• The method showed universality by successfully diagnosing faults in a different rotor equipment.

Conclusions:

• Direct classification of CWTS using CNNs is an effective approach for rotating machinery fault diagnosis.
• This method retains comprehensive information from vibration signals, overcoming limitations of feature extraction.
• The developed technique offers a reliable and generalizable solution for ensuring machinery health.