Development of a state prediction model to aid decision making in condition based maintenance

Condition monitoring and fault diagnosis for operational equipment are developing andshowing their potential for enhancing the effectiveness and efficiency of maintenancemanagement, including maintenance decision-making. In this thesis, our aim is tomodel the condition of equipment items subject to condition-monitoring in order toprovide a quantitative measure to aid maintenance decision-making. A key ingredienttowards dealing with the modelling work is to define the state or condition of theequipment with an appropriate measure and the observed condition monitoring may bea function of the state or condition of the operational equipment concerned. This leadsto the two elements that are important in our modelling development; the need todevelop a model that describes the system condition subject to its monitoring data and adecision model that is based upon the predicted system condition.A quantification of the system condition in this thesis is modelled using either discreteor continuous measures. In the case of a discrete state space, this thesis presents detailsof how the initiation of a random defect can be identified. In the case of a continuousstate space, two approaches, which were used to identify the system condition, arediscussed. The first is adopted from the concept of the conditional residual time andsecondly, a wear process determined from a beta distribution. In developing thesemodels, we used vibration and oil analysis data. Note that understanding, manipulatingand analysing of the data played an important role in this thesis. This is needed not onlyfor model development, but also for validating the model. Methods for estimatingmodel parameters are discussed in detail. In addition, since the models presented aregenerally beyond the scope for analytical solutions, two numerical approximationmethods are proposed. Simple decision models, which minimize the expected cost perunit time over a time interval between the current monitoring time and the nextmonitoring time, are shown. Numerical examples to demonstrate the modelling ideasare also illustrated throughout the thesis.

Level dc:type.qualificationlevel

Doctoral (Level 8)

Year dc:date.issued

2007

Author dc:creator

• Hussin, B

Language dc:language

en

Identifier

oai:salford-repository.worktribe.com:1337098

OAI identifier oai:identifier

oai:salford-repository.worktribe.com:1337098