| Abstract: | This paper considers a repairable machine that may fail or sufferbreakdown many times during the course of its service lifetime,and is inspected for visible faults at intervals. The delay-timeconcept of Christer & Waller provides a means of modellingthe behaviour of the system, and predicting such useful quantitiesas reliability and cost, under various putative inspection regimes. Hitherto, model parameters have been estimated mainly from subjectivedata. In this paper, we show that it is both theoretically andpractically possible to estimate model parameters, and makeuseful predictions, purely from objective data, i.e. the historyof breakdown times and the findings of inspections. Model parameters are fitted by the method of maximum likelihood,and selection of the 'best' model made using the Akaike informationcriterion (AIC). Initially, Monte-Carlo studies were made, andshowed that the p r d u r e did enable unbiased and asymptoticallyaccurate estimates of model parameters to be recovered fromdata. Manual records of inspections and failures of a sampleof hospital infusion pumps were then analysed, and values ofmodel parameters estimated. Tests of fit were derived and carriedout. Finally, the reliability of infusion-pump components underdifferent inspection intervals was derived from the delay-timemodel with 95% confidence limits, as a demonstration that themethod does indeed provide a practical tool for optimizing inspectionpolicies. The practical details of the relevant computations are givenin some detail throughout, to enable other workers to followour procedure. |
