Abstract:
This implementation report presents a maintenance scheduling tool for Kaplan turbines.
Therein is the system decomposition of the Kaplan turbine and a Failure Modes, Effects, Criticality and Diagnostic Analysis of its components. As a result, a process map was developed in form of an adjacency matrix and a figure of component interconnectivity. Weibull analysis was invoked on the maintenance data and a table that shows the component details discovered. The shape parameters which indicate the stage of the turbine components was developed, thus guidance for the type of maintenance to give to a component. Guide vanes were discovered as the most vulnerable component with a repair frequency of nearly a month, a shape parameter of 0.6141 and a scale parameter of 10 days, implying that they are still in the burn in stage, and possible remedies given so as to tame the water quality. Furthermore, a diagnostic tool was developed using the Bayesian Networks and Hidden Markov chain. The model was established in terms of transition and emission probabilities, which were given in terms of matrices. Program Evaluation and Review Technique (PERT) analysis was used to obtain maintenance project duration for the critical path for the maintenance of each of the components. Later this knapsack problem was be fed into a Genetic Algorithm to optimize the maintenance schedule, putting into consideration the maintenance window together with the flows and power prediction peak and off-peak periods. The intelligent optimization models were be developed in MATLAB and thereafter the algorithms tested on a case of Bujagali hydropower plant.
Keywords: vertical Kaplan turbines, fault diagnosis, Bayesian networks, Hidden Markov chains, maintenance scheduling
