A Bayesian inference-based approach for performance prognostics towards uncertainty quantification and its...

Area of Science:

• Mechanical Engineering
• Data Science
• Reliability Engineering

Background:

• Marine diesel engines require robust performance prognostics to ensure operational reliability and safety.
• Uncertainty in performance predictions can lead to suboptimal maintenance scheduling and potential failures.
• Existing condition monitoring tools may lack the accuracy needed for complex engine systems.

Purpose of the Study:

• To introduce a Bayesian analysis framework for marine diesel engine performance prognostics.
• To develop and compare Bayesian neural networks for health monitoring and Bayesian logistic regression for run-to-failure analysis.
• To validate the proposed framework using real-world engine operational data.

Main Methods:

• Utilized Bayesian neural networks for health monitoring and Bayesian logistic regression for run-to-failure quantification.
• Employed Variational Inference and Markov Chain Monte Carlo algorithms for model parameter learning.
• Selected instantaneous angular speed signals for indirect prediction of indicated mean effective pressure.

Main Results:

• The proposed Bayesian framework demonstrated superior performance compared to conventional methods.
• The Bayesian models effectively addressed uncertainty in inferences and results.
• Instantaneous angular speed signals proved valuable for indirect performance assessment.

Conclusions:

• The developed Bayesian framework offers a reliable approach for marine diesel engine performance prognostics.
• The method shows significant potential for application as an online condition monitoring tool.
• Bayesian analysis enhances the accuracy and reliability of engine health monitoring and failure prediction.