Accommodating maintenance in prognostics

  • Omer Ahmed Khan Panni

Student thesis: Doctoral Thesis

Abstract

Steam turbines are an important asset of nuclear power plants, and are required tooperate reliably and efficiently. Unplanned outages have a significant impact on theability of the plant to generate electricity. Therefore, condition-based maintenance (CBM)can be used for predictive and proactive maintenance to avoid unplanned outages whilereducing operating costs and increasing the reliability and availability of the plant. InCBM, the information gathered can be interpreted for prognostics (the prediction offailure time or remaining useful life (RUL)).The aim of this project was to address two areas of challenges in prognostics, theselection of predictive technique and accommodation of post-maintenance effects, toimprove the efficacy of prognostics. The selection of an appropriate predictive algorithmis a key activity for an effective development of prognostics. In this research, a formalapproach for the evaluation and selection of predictive techniques is developed tofacilitate a methodic selection process of predictive techniques by engineering experts.This approach is then implemented for a case study provided by the engineering experts.Therefore, as a result of formal evaluation, a probabilistic technique the Bayesian LinearRegression (BLR) and a non-probabilistic technique the Support Vector Regression (SVR)were selected for prognostics implementation.In this project, the knowledge of prognostics implementation is extended by includingpost maintenance affects into prognostics. Maintenance aims to restore a machine into astate where it is safe and reliable to operate while recovering the health of the machine.However, such activities result in introduction of uncertainties that are associated withpredictions due to deviations in degradation model. Thus, affecting accuracy and efficacyof predictions. Therefore, such vulnerabilities must be addressed by incorporating theinformation from maintenance events for accurate and reliable predictions. This thesispresents two frameworks which are adapted for probabilistic and non-probabilisticprognostic techniques to accommodate maintenance. Two case studies: a real-world casestudy from a nuclear power plant in the UK and a synthetic case study which wasgenerated based on the characteristics of a real-world case study are used for theimplementation and validation of the frameworks. The results of the implementationhold a promise for predicting remaining useful life while accommodating maintenancerepairs. Therefore, ensuring increased asset availability with higher reliability,maintenance cost effectiveness and operational safety.
Date of Award19 Oct 2020
Original languageEnglish
Awarding Institution
  • University Of Strathclyde
SponsorsUniversity of Strathclyde, EPSRC (Engineering and Physical Sciences Research Council) & Rolls-Royce PLC
SupervisorGraeme West (Supervisor) & Stephen McArthur (Supervisor)

Cite this

'