Joint Learning of Failure Mode Recognition and Prognostics for Degradation Processes

Area of Science:

• Engineering
• Computer Science
• Data Science

Background:

• Accurate prognostics health management (PHM) is crucial for preventing unexpected failures in manufacturing systems.
• Existing methods often treat failure mode recognition and remaining useful lifetime (RUL) prediction independently, limiting accuracy.
• RUL prediction is highly dependent on specific failure modes due to distinct sensor signal degradation patterns.

Purpose of the Study:

• To propose a novel joint learning model for simultaneous failure mode recognition and RUL prediction.
• To leverage interpretable degradation features extracted from multiple sensor signals for improved PHM.
• To enhance the accuracy of RUL prediction by incorporating failure mode information.

Main Methods:

• Developed a joint learning model integrating multiple sensor signals for degradation processes.
• Extracted interpretable features considering degradation mechanisms as input for a deep neural network.
• Trained the model using historical unit data, including sensor signals, failure times, and failure modes.

Main Results:

• The joint learning model successfully performs failure mode recognition and RUL prediction simultaneously.
• The model effectively characterizes complex relationships between features, RUL, and failure modes.
• Demonstrated effectiveness through a case study on aircraft gas turbine engine degradation.

Conclusions:

• Jointly learning failure mode recognition and RUL prediction significantly improves PHM accuracy.
• The proposed data-driven neural network approach is flexible and applicable to complex manufacturing systems.
• This method offers a robust solution for predicting unit RUL and identifying failure modes in real-world applications.