Statistical models for expert judgement and wear prediction: Dissertation

Urho Pulkkinen

Research output: ThesisDissertationCollection of Articles


This thesis studies the statistical analysis of expert judgements and prediction of wear. The point of view adopted is the one of information theory and Bayesian statistics. A general Bayesian framework for analyzing both the expert judgements and wear prediction is presented. Information theoretic interpretations are given for some averaging techniques used in the determination of concensus distributions. Further, information theoretic models are compared with a Bayesian model. The general Bayesian framework is then applied in analyzing expert judgements based on ordinal comparisons. In this context, the value of information lost in the ordinal comparison process is analyzed by applying decision theoretic concepts. As a generalization of the Bayesian framework, stochastic filtering models for wear prediction are formulated. These models utilize the information from condition monitoring measurements in updating the residual life distribution of mechanical components. Finally, the application of stochastic control models in optimizing operational strategies for inspected components are studied. Monte-Carlo simulation methods, such as the Gibbs sampler and the stochastic quasi-gradient method, are applied in the determination of posterior distributions and in the solution of stochastic optimization problems.
Original languageEnglish
QualificationDoctor Degree
Awarding Institution
  • Helsinki University of Technology
  • Hämäläinen, Raimo, Supervisor, External person
Award date27 May 1994
Place of PublicationEspoo
Print ISBNs951-38-4419-6
Publication statusPublished - 1994
MoE publication typeG5 Doctoral dissertation (article)


  • statistical models
  • bayes theorem
  • comparison
  • modelling
  • stochastic processes
  • filtering
  • optimization
  • theses


Dive into the research topics of 'Statistical models for expert judgement and wear prediction: Dissertation'. Together they form a unique fingerprint.

Cite this