Interactive Prognosis Framework Between Deep Learning and a Stochastic Process Model for Remaining Useful Life...

Area of Science:

• Reliability Engineering
• Machine Learning
• Data Science

Background:

• Accurate remaining useful life (RUL) prediction is crucial for degraded systems, especially in the big data era.
• Existing methods often suffer from a mismatch between deep learning-based health indicator (HI) construction and stochastic degradation modeling.
• This defect significantly impacts the accuracy of RUL predictions.

Purpose of the Study:

• To propose an interactive prognosis framework that bridges the gap between deep learning and stochastic process models for RUL prediction.
• To enhance the accuracy of RUL prediction by addressing the matching defect in conventional approaches.

Main Methods:

• Utilized stacked contractive autoencoders for unsupervised health indicator (HI) construction by fusing multi-sensor data.
• Introduced an exponential-like degradation model to capture the nonlinear characteristics of the constructed HI.
• Derived theoretical expressions for prediction results using the first hitting time concept and optimized an integrated objective function via gradient descent.

Main Results:

• The proposed interactive prognosis framework successfully integrates HI construction and degradation modeling.
• The method generates the HI for field systems and provides the probability density function (pdf) of the predicted RUL.
• Validation through two case studies on turbofan engines demonstrated the effectiveness and superiority of the proposed method.

Conclusions:

• The interactive prognosis framework offers a robust solution for RUL prediction by overcoming the HI-degradation model matching defect.
• The approach provides accurate RUL predictions and uncertainty quantification for degraded systems.
• This method holds significant potential for applications in predictive maintenance and system health management.