Managing the growing amount of dynamic sensor data in power plants

摘要

The number of sensors in power plants is increasing every year. The processing power of CPUs, hard disk storage capacities and transfer band widths are growing as well, but the amount of data is outnumbering this growth, especially when high resolution dynamic data is generated for an ever increasing number of sensors. This data includes for examples dynamic data with high sampling rates (50 kHz), spectra with a high number of spectral lines (6400). All this data allows for an optimized operation of the power plant by including them in the closed loop control. Apart from that, this data also needs to be transferred and stored such that it can be visualized and analyzed by engineers worldwide far away from the actual power plant. This analysis eases understanding of events and enables a comparison of machines and their optimization across a fleet. This work describes different strategies to deal with the tradeoff between storing and transferring all this data and the limited storage space and transfer rates. For example, with 32 channels of dynamic pressure data at a sampling rate of 25.6 kHz and single precision floating point accuracy, one gets 283 GB of raw time data and 354 GB of spectra data per day. With a hard disk of 8 TB this allows to store data for just about 13 days resp. to store one year one needs 227 TB. This not only demands many resources at the data source, but also at the sink, i.e. the processing of the data for visualization and analysis. KData Storage;;Signal Analysis;;Statistics;;Combustion Dynamics;;Humming;;Pulsation;;Gas Turbine Control and Monitoring