A deep learning-based prognostic approach for predicting turbofan engine degradation and remaining useful life

Area of Science:

• Aerospace Engineering
• Artificial Intelligence
• Mechanical Engineering

Background:

• Component degradation in turbofan engines poses significant risks to aviation safety.
• Accurate Remaining Useful Life (RUL) prediction is vital for effective prognostics and health management (PHM).
• Existing methods require enhancement for improved RUL prediction accuracy.

Purpose of the Study:

• To propose a novel deep learning model for accurate RUL prediction of turbofan engines.
• To evaluate the proposed model's performance on the CMAPSS benchmark dataset (FD001 and FD003).
• To demonstrate the model's superiority over existing state-of-the-art methods.

Main Methods:

• A hybrid Convolutional Autoencoder and Attention-based LSTM (CAELSTM) model was developed.
• Piecewise linear degradation modeling and data preprocessing were applied to the CMAPSS dataset.
• An autoencoder, attention-based LSTM, and fully connected layers were utilized for feature extraction and RUL prediction.

Main Results:

• The CAELSTM model achieved superior RUL prediction performance on FD001 and FD003 sub-datasets.
• Achieved Root Mean Square Error (RMSE) of 14.44 (FD001) and 13.40 (FD003).
• Mean Absolute Error (MAE) of 10.49 (FD001) and 10.68 (FD003), and scores of 282.38 (FD001) and 264.47 (FD003) were recorded.

Conclusions:

• The proposed CAELSTM model demonstrates significant effectiveness and superiority for turbofan engine RUL prediction.
• This advancement offers a dependable tool for predictive maintenance in aerospace, enhancing aviation safety.
• The model shows great promise for improving prognostics and health management systems.