Anomaly indicators for Kaplan turbine components based on patterns of normal behavior

Abstract

This paper describes and proposes some indicators for continuous monitoring of anomalous conditions in the hydraulic system of a Kaplan turbine using SCADA data. The indicators are based on significant deviations between the estimated values for key variables describing the current working conditions of the components at the plant, and those actually observed. This monitoring strategy requires models describing the expected values for variables through the whole range of possible working conditions of the monitored components. These models are normal behavior models able to characterize the typical relationships between a set of variables used as inputs to the models and the corresponding output of a target variable whose expected value has to be predicted. The criteria to select the variables to use in the models are based on the physical working principles of the component. The paper is focused on models of normal behavior applied to a real case of condition monitoring of a Kaplan turbine regulating mechanism.

Anomaly indicators for Kaplan turbine components based on patterns of normal behavior