Improved sand cat swarm optimization algorithm assisted GraphSAGE-GRU for remaining useful life of engine

Area of Science:

• Mechanical Engineering
• Artificial Intelligence
• Data Science

Background:

• Deep learning methods for Remaining Useful Life (RUL) prediction offer powerful data processing but often overlook parameter interdependencies in non-Euclidean spaces during engine degradation.
• Existing deep learning approaches may not fully capture the complex relationships within operational parameters, limiting prediction accuracy.

Purpose of the Study:

• To propose an improved sand cat swarm optimization-assisted GraphSage-GRU (ISCSO-GraphSage-GRU) model for accurate engine RUL prediction.
• To address the limitations of current deep learning methods by incorporating parameter interdependencies in non-Euclidean spaces.

Main Methods:

• Utilized Maximum Information Coefficient (MIC) to identify and describe interdependent relationships among measured engine parameters.
• Developed an ISCSO-GraphSage-GRU model, feeding graph data derived from parameter interdependencies into GraphSage-GRU.
• Enhanced the sand cat swarm optimization (SCSO) with tent mapping for population initialization and a novel adaptive approach to improve exploration and exploitation.

Main Results:

• The ISCSO-GraphSage-GRU model demonstrated high effectiveness and advanced performance on the CMAPSS dataset.
• Achieved a coefficient of determination (R²) greater than 0.99, indicating a strong fit of the model to the data.
• Obtained a Root Mean Square Error (RMSE) of less than 6, signifying high prediction precision.

Conclusions:

• The proposed ISCSO-GraphSage-GRU model successfully overcomes the limitations of existing deep learning methods in engine RUL prediction.
• The integration of MIC for parameter interdependency analysis and ISCSO for model optimization leads to superior predictive performance.
• The study validates the advanced capabilities of the developed model for accurate and reliable engine health monitoring.