Thesis

Incorporating expert judgement into condition based maintenance decision support using a coupled hidden markov model and a partially observable markov decision process

Downloadable Content

Download PDF
Creator
Rights statement
Awarding institution
  • University of Strathclyde
Date of award
  • 2012
Thesis identifier
  • T13448
Qualification Level
Qualification Name
Department, School or Faculty
Abstract
  • Preventive maintenance consists of activities performed to maintain a system in a satisfactory functional condition. Condition Based Maintenance (CBM) aims to reduce the cost of preventive maintenance by supporting decisions on performing maintenance actions, based on information reflecting a system's health condition. In practice, the condition related information can be obtained in various ways, including continuous condition monitoring performed by sensors, or subjective assessment performed by humans. An experienced engineer might provide such subjective assessment by visually inspecting a system, or by interpreting the data collected by condition monitoring devices, and hence give an "expert judgement" on the state of the system. There is limited academic literature on the development of CBM models incorporating expert judgement. This research aims to reduce this gap by developing models that formally incorporate expert judgement into the CBM decision process. A Coupled Hidden Markov Model is proposed to model the evolutionary relationship between expert judgement and the true deterioration state of a system. This model is used to estimate the underlying condition of the system and predict the remaining time to failure. A training algorithm is developed to support model parameter estimation. The algorithm's performance is evaluated with respect to the number of expert judgements and initial settings of model parameters. A decision-making problem is formulated to account for the use of expert judgement in selecting maintenance actions in light of the physical investigation of the system's condition. A Partially Observable Markov Decision Process is proposed to recommend the most cost-effective decisions on inspection choice and maintenance action in two consecutive steps. An approximate method is developed to solve the proposed decision optimisation model and obtain the optimal policy. The sensitivity of the optimal policy is evaluated with respect to model parameters settings, such as the accuracy of the expert judgement.
Resource Type
Note
  • Strathclyde theses - ask staff. Thesis no. : T13448
DOI
Date Created
  • 2012
Former identifier
  • 991378

Relations

Items

Thumbnail Title Date Uploaded Visibility Actions
file details PDF of thesis T13448 2021-07-02 Public Download