Predicting Failures for Repairable System Subjected to Imperfect Maintenance

The purpose of this project is to develop a reliability model which results from
reliability analysis conducted on repairable system subjected to imperfect
maintenance. Hence, in order to perform the reliability analysis, field data from
actual equipment failure were gathered and analyzed. In this project, the equipment
selected was the centrifugal pump used in one of the petrochemical plants. Various
stages had been conducted in order to achieve the objectives of the project. This
includes data screening and analysis, determination of failure distribution as well as
the maintenance effectiveness which denoted by q. All of these phases were
performed by using the reliability software, Weibull ++7. The data analysis showed
that the failure data displayed Weibull distribution while q value indicated the
Generalized Renewal Process (GRP) is the most applicable probabilistic models that
characterized the failure data. Thus, the reliability model was developed by using
GRP model of Type I and Type II. The comparison between both models was
conducted to select the suitable model to be used in developing the reliability model.
Based on the likelihood value (LV), GRP model Type I was selected as it possessed
higher LV and this model was used to predict the future failures of the system.
Evaluation phase was conducted to verify that GRP model Type I was the most
suitable model which fits best the failure data. In this phase, the reliability model was
developed by using other probabilistic models such as Renewal Process (RP) and
Non-Homogeneous Poisson Process (NHPP). The LV were compared which resulted
in GRP model Type I produced the highest LV. Finally, the model was validated by
using reliability models developed based on the different duration of operation days
which were 1500 and 2000 operation days, respectively. The expected cumulative
numbers of failures calculated by both models were then compared with the actual
cumulative number of failures obtained from the model developed using 3000
operation days. Based on the comparison, both models produced similar values with
the actual failure data. Hence, the developed reliability model could be used to
predict the next failure of the system. It is hoped that this project and report could be
used as a reference for further research and study.