Use this URL to cite or link to this record in EThOS:
Title: Some aspects of Weibull analyses of reliability data
Author: Griffin, R. M.
Awarding Body: University College of Swansea
Current Institution: Swansea University
Date of Award: 1995
Availability of Full Text:
Access from EThOS:
The thesis presents research into aspects of reliability based on monitoring the operation and management of large fleets of vehicles and equipment by a United Kingdom Government Department. A description of the Department and the results of two studies into the organisation are presented: on the data capture system used by the organisation, and on a sample of data from the organisation's data centre. The results of an equipment census are reported showing the effects of calculating mean time to failure on fails only and fails and survivor data are contrasted. Examination of the methods of analysis within the organisation have led to the need for comparison of the Weibull and negative exponential distributions in terms of their efficiency in producing parameter estimates from failure information only and fails and survivor data. Simulation experiments are described which illustrate this comparison in the presence of multiple censoring. The comparisons indicated that for complete samples the Weibull was more efficient, generally providing higher quality estimates in a shorter time, although there were circumstances when there was less to choose between the models. The studies indicated that manual transcription of data into the data capture system, sometimes repeated at differing locations led to a number of data errors; misplacement of decimal point, arbitrary zeroing of indicators (zeroing), duplicate data entry, enhancement of indicator reading (run-on), empirical conversion and estimation of data (mis-read). A series of simulations are described which investigate the effects of a corruption factor ρ for 0.00 ≤ ρ ≤ 0.10, in the absence of censoring for; the mis-placed decimal point, mis-read, run-on and zeroing corruption scenarios. Introduction of type I censoring allowed further simulation experiments, designed to determine the effects of mis-classification of fails and censored timings for a corruption factor ρ for 0.00 ≤ ρ ≤ 0.10. Profile inaccuracy was used to calculate values of for 0.00 ≤ ρ ≤ 0.10, for three corruption scenarios, which gave good agreement with values of β and θ obtained by simulation. A theory is presented which establishes the variance covariance structure of maximum likelihood estimators with those generated by simulation. A worked example is provided to compare with and to correct E[Iθθ] in table 3 and to enhance table 1 of Watkins and Griffin (1994).
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available