Rotating machinery reliability

Rotating machinery is a major contributor to process plant downtime, hence, predicting the reliability performance of these complex systems becomes an important task in plant availability assessments. Currently the models employed are relatively elementary so the objective of this research has been to investigate alternatives which offer improvements particularly for assessing the effect of High-Impact, Low-probability (HILP) events. Data are necessary in puts to availability models so a significant part of the researchh as focused on the analysis of reliability and maintainability data. Generic and in-house sources or rotating machinery failure and repair data have been used. Gas turbines with power outputs less than 30MW, which are employed as drivers for a variety of process applications, provide the main source of the data analysed. In this area the popular assumption of increasing rates of failure with equipment age is not supported. Indications of increasing failure rates (IFR), where they do exist, are quite weak and it is evident that gas turbine systems generally show slowly decreasing failure rates (DFR) with time in service. Constant failure rates for gas turbine systems is therefore a reasonable assumption for availability studies. The majority of the data analysed however have limitations for predicting the likely failure rate for a specific type of gas turbine in a well-specified application. The problems of maintainability data are less acute although it is evident that the assumption of a single constant repair rate for complex equipment such as a gas turbine is over simplistic. Repair time distributions are strongly skewed, however, when repair times are partitioned into 3 or more ranges the assumption of constant repair rates appears viable. The alternative model proposed is based on the assumption of 3 exponentially-distributed failure/repair modes. This model predicted significantly lower availabilities than single failure/repair mode models which underestimates the importance of the High-Impact-Low- Probability outages. However, the model was found to present problems for the application in Markovian and simulation analyses for more complex configurations (for example, 2-out-of-3 systems) because, inter-alia, of the complexity of'construdting representative State Transition Diagrams. The alternative of employing fault trees proved significantly more tractable. It was shown to give the same numerical output as obtained by Markov analysis and has many other advantages. These include its hierarchical constructipp which facilitates the representation of complex system logic, the scope for decomposing a complex system into sub-trees by the use of transfer-in, transfer-out gates, the ability to introduce delays, sequential events and the selective use of simulation in important sub-trees.