Generalised Proportional Intensities Models (gpim) for Reliability Analysis

Stochastic models are developed for the reliability analysis of repairable systems, based upon the non-homogenous Poisson process, as recommended by Ascher and Feingold (1984). In particular, the proportional intensities model (PIM) introduced by Cox (1972), is extended and modified for this purpose. The suitability of the PIM is demonstrated on three groups of hypothetical data sets from the first of these two books. Having identified potential benefits from this approach, the PIM is extended and a new class of generalized proportional intensities models (GPIM) is introduced. These allow for modelling realistic failure patterns by including preventive maintenance activities and explanatory variables to gain valuable insights about cost effective maintenance and replacement strategies. This thesis presents the algebraic theory and develops several variations of the basic GPIM. It also comments on similarities and differences between these and other proposed models for complex repairable systems. Practical applications of the GPIM models are demonstrated on published data sets that were collected from petroleum refineries. We also explore a new area by fitting the models to British Petroleum gas turbine maintenance data from an oil platform in the North Sea, using the Fortran 95 programming language and Mathcad mathematical software. Finally, following the in-depth analysis of the refinery pump data sets and oil gas turbines engine maintenance data, the GPIM model is simulated to minimize the expected cost per unit time over an illustrative fixed horizon of ten years, in order to determine optimal PM strategies. Along the way, weaknesses in current data selection systems are identified, which allow us to make general recommendations for recording xi maintenance history data. This research was motivated by the Nigerian oil industry where system availability is poor and few maintenance records are kept.