A new ensemble residual convolutional neural network for remaining useful life estimation

Area of Science:

• Artificial Intelligence
• Machine Learning
• Deep Learning

Background:

• Remaining Useful Life (RUL) estimation is crucial for prognostic health management (PHM) in modern industry.
• Data-driven RUL approaches are gaining prominence with smart manufacturing advancements.
• Classical deep learning methods face vanishing/exploding gradient issues.

Purpose of the Study:

• To propose a novel Residual Convolutional Neural Network (ResCNN) for enhanced RUL estimation.
• To address the vanishing/exploding gradient problem in deep learning models for RUL prediction.
• To improve the accuracy and reliability of RUL predictions through an ensemble method.

Main Methods:

• Developed a ResCNN incorporating residual blocks with shortcut connections to mitigate gradient problems.
• Enhanced the ResCNN model using a k-fold ensemble technique.
• Evaluated the proposed model on the NASA C-MAPSS dataset.

Main Results:

• The ensemble ResCNN demonstrated significant improvements in both the mean and standard deviation of RUL predictions.
• Achieved state-of-the-art performance compared to various machine learning and deep learning methods.
• Outperformed established models like Multilayer Perceptron, SVM, DBN, and LSTM.

Conclusions:

• The proposed ensemble ResCNN effectively overcomes gradient issues inherent in deep learning.
• This approach offers superior accuracy and reliability for RUL estimation in PHM systems.
• The method shows significant potential for industrial applications in predictive maintenance.