A novel condition monitoring methodology based on neural network of pump-turbines with extended operating range

Abstract

Due to the entrance of new renewable energies, water-storage energy has to be regulated more frequently to keep the stability of power grid. Consequently, pump-turbines have to work under offdesign conditions more than before, which will cause more damage and decrease their useful life. Advanced monitoring methodologies that can balance the degradation of machine and revenues to the power plant has been required. To develop an innovative condition monitoring approach, vibration data was collected from different components of a pump-turbine which is running in an extended operating range. The consequences of operating range extension on the vibration of the pump-turbine have been studied by analysing the vibration signatures. The changing rule of the vibration behavior of the machine with the operating parameters has been obtained. An artificial neural network based model has been applied to build an autoregressive normal behavior model. The results indicated that the normal behavior model based on multi-layer neural net has the ability to predict the vibration characteristics of the machine in different operating conditions. This monitoring method can be adapted to the similar type of hydraulic turbine units.