Monitoramento de condição de uma unidade geradora hidrelétrica

Abstract

Hydroelectric power plants are responsible for most of the energy generation of Brazils electrical grid. Monitoring and attesting the proper functioning of the turbine-generator units is essential, especially in periods of drought, when the operation conditions might be atypical. In that perspective, a development and application of techniques is presented aiming a broad condition monitoring of a generator unit, using multivariate statistical process control tools, such as Hotellings 2 chart, SPE and combined indexes, besides PCA and contribution plots. This unit is part of a hydroelectric power plant that has particular modes of operation, which include constant variations of setpoint, causing non-stationarity conditions. Therefore, a methodology is proposed for the analysis and monitoring of pertinent measurements using variable transformations, recursive variance estimation for identifying operational modes, mean filters at the desing step and a procedure for mean adaptation for just-in-time monitoring. The algorithms are developed in Matlab, tested with historical data from the plant, along with PCA analysis for identifying the variables that most likely caused the alarms generate by the control charts. The methodology is developed aiming further implementation on a PIMS system, installed in the plants site, to integrate monitoring solutions for the processes. The results show that the methodology allows monitoring a process with different operational points and modes, even if some of them have not been observed in the retrospective data analysis