Bayesian Architecture for Predictive Monitoring of Unbalance Faults in a Turbine Rotor-Bearing System.

Area of Science:

• Mechanical Engineering
• Reliability Engineering
• Condition Monitoring

Background:

• Unbalance faults are a primary cause of failures in rotary systems, necessitating effective monitoring for predictive maintenance.
• Conventional models often assume time-invariant bearing parameters, which is unrealistic due to operational and environmental changes.
• Accurate assessment of bearing characteristics is crucial for reliable fault detection.

Purpose of the Study:

• To develop a novel architecture for monitoring and predicting unbalance faults in rotor systems.
• To incorporate dynamic bearing parameters that adapt to changing operating conditions.
• To enhance the reliability of mathematical models for continuous fault assessment.

Main Methods:

• A Bayesian inference framework utilizing Markov Chain Monte Carlo (MCMC) sampling.
• The Metropolis algorithm was employed for systematic evaluation of parameter values.
• A dual-MCMC loop approach was implemented for comprehensive parameter space exploration.

Main Results:

• Accurate estimation of dynamic bearing parameters was achieved.
• The novel framework improved unbalance assessment by up to 74.48% in residual error reduction compared to fixed-parameter models.
• Validated the effectiveness of the Bayesian framework for predictive monitoring.

Conclusions:

• The proposed dynamic parameter approach offers a more reliable method for unbalance fault monitoring.
• This framework supports advanced predictive maintenance strategies in rotor systems.
• The study demonstrates significant improvements in fault assessment accuracy.