Sensor set optimization by functional model and Bayesian network for fault diagnosis of turbine generator lubrication system

Turbine generator lubrication system, providing oil at an acceptable temperature, pressure, quantity and cleanliness to the bearings, is one of the safety-critical auxiliary systems in the power plant operation. The fault diagnosis capability of the system by utilizing relevant sensors is highly important but can be a great challenge since it comprises of numerous elements including tank, pump, filter, and so on. In this paper, an efficient method is presented to design optimum sensor set for the fault diagnosis of turbine generator lubrication system. Toward this objective, a functional model is created to represent the system and elements by the causal links between flow properties. However, since the concept of link strength is hard to interpret in practice, a novel approach is introduced to determine these inversely by applying parameter learning of Bayesian network with the empirical knowledge on the flow differential in each element. Using the constructed model, faults are simulated for selected failure modes to obtain propagation table. Optimum sensor set design is explored by genetic algorithm with the objective to identify all the failure modes with minimum number of sensors and types.