A CASE STUDY OF RELIABILITY ASSESSMENT FOR CENTRIFUGAL PUMPS IN A PETROCHEMICAL PLANT

• Main Authors: Muhammad, Masdi, Abd Majid, M Amin, Ibrahim, Nurul Akma
• Format: Conference or Workshop Item
• Published: 2009
• Subjects: TJ Mechanical engineering and machinery
• Online Access: http://eprints.utp.edu.my/3723/1/398-404.pdf; http://eprints.utp.edu.my/3723/

Centrifugal pumps are widely used in petrochemical industry and in some instances, the number of pumps used could easily amount to hundreds of pumps in a typical petrochemical plant. Consequently, the reliability of these pumps essentially translates into stable and reliable plant operation as the pumps performances are critical in ensuring continuous plant productivity. Reliability assessment for repairable equipment, which in this case centrifugal pumps, is highly dependent upon the assumption of the state after each repair. The post repair states can be categorized into three different states namely, ‘as good as new’, ‘as bad as old’ and the states in between. In practice, however, the usual state of equipment after repair follows the state of ‘better than old but worse than new’ which lies somewhere in between the two extremes. This paper focuses on the reliability assessment of the centrifugal pumps at a refinery plant that has been in operation for more than 10 years using a more robust process called generalized renewal process (GRP). This process has been proposed to model not only the ‘inbetween’ states but also the two extreme post repair states. A case study utilizing centrifugal pump failure data is used as a comparative appraisal of reliability assessment between GRP, perfect renewal process (PRP) and non-homogenous Poisson process (NHPP). The underlying distribution for time to first failure for these pumps is assumed to follow the two-parameter Weibull distribution and the parameters for the models are estimated using maximum likelihood estimation (MLE). The GRP solution based on the case study showed better description of the failure distribution even with limited available failure data in contrast with other assumptions as indicated by the likelihood values.