A negative correlation ensemble transfer learning method for fault diagnosis based on convolutional neural network.

Area of Science:

• Engineering
• Computer Science
• Artificial Intelligence

Background:

• Smart manufacturing relies on data-driven fault diagnosis.
• Deep learning (DL) shows promise but faces challenges in model depth and generalization.
• Existing DL models struggle with diverse datasets due to the no-free-lunch theorem.

Purpose of the Study:

• To propose a novel negative correlation ensemble transfer learning (NCTE) method.
• To address the limitations of shallow DL models and poor generalization in fault diagnosis.
• To enhance the accuracy and robustness of fault diagnosis systems.

Main Methods:

• Developed a 50-layer deep learning structure using transfer learning with ResNet-50.
• Employed negative correlation learning (NCL) to cooperatively train fully-connected layers and softmax classifiers.
• Utilized cross-validation for hyper-parameter optimization of the NCTE method.

Main Results:

• The proposed NCTE method achieved a prediction accuracy of 98.73% on the KAT Bearing Dataset.
• NCTE demonstrated superior performance compared to other existing machine learning and deep learning methods.
• The study validates the effectiveness of ensemble transfer learning in deep learning-based fault diagnosis.

Conclusions:

• The NCTE method offers a robust solution for data-driven fault diagnosis in smart manufacturing.
• NCTE effectively overcomes the generalization limitations of individual deep learning models.
• This research contributes to advancing the field of intelligent fault diagnosis systems.