Multi-Scale Temporal-Spatial Feature-Based Hybrid Deep Neural Network for Remaining Useful Life Prediction of...

Area of Science:

• Aerospace Engineering
• Mechanical Engineering
• Data Science

Background:

• Remaining Useful Life (RUL) prediction is vital for aero-engine maintenance and longevity.
• Existing deep neural networks for RUL prediction often neglect data point importance and sensor interdependencies.

Purpose of the Study:

• To develop an advanced deep neural network for enhanced RUL prediction in aero-engines.
• To address limitations in current RUL prediction models by incorporating multiscale temporal-spatial features.

Main Methods:

• Proposing the Multi-Scale Temporal-Spatial feature-based hybrid Deep neural Network (MSTSDN).
• Extracting multiscale features from raw multi-sensor monitoring data.
• Evaluating MSTSDN on C-MAPSS and N-CMAPSS aero-engine datasets.

Main Results:

• MSTSDN demonstrated superior accuracy and timeliness in RUL predictions compared to 12 existing models on the C-MAPSS dataset.
• The model showed effectiveness across diverse operational conditions and fault modes.
• MSTSDN successfully tracked and fitted actual RUL during engine degradation on the N-CMAPSS dataset.

Conclusions:

• The proposed MSTSDN significantly enhances RUL prediction performance for aero-engines.
• MSTSDN offers a robust solution for complex degradation monitoring and maintenance planning.
• This model provides a valuable advancement for predictive maintenance in aerospace applications.