Abstract
In Japan, gas turbine combined cycle (GTCC) power plants are currently operated under rapid start-up, shutdown, and partial-load conditions to maintain a stable power supply, suppressing the fluctuation of electric power due to renewable energy sources such as solar and wind power. The off-design condition may cause unsteady and unstable flows in the compressor stage, increasing mechanical stresses on the blades. To reveal the critical flow behavior and avoid blade failure, an innovative method that can clarify the flow physics in these systems is urgently required.
This study proposes a classification method for time-series big data obtained from full-annulus flow computations under various flow conditions assisted by the self-organizing map (SOM) method, which is a popular machine-learning technique. First, full-annulus unsteady flow computations under various flow conditions were conducted for an actual compressor in an industrial gas turbine at a GTCC power plant. The inlet guide vane (IGV) angle was set to various values that corresponded to the full-load operation and transitional flow rate conditions during partial-load operation. The time-series data of the load on the blades in the various cases were obtained from the computations. All time-series data were converted into frequency data using fast Fourier transform (FFT) and classified by the SOM. As a result, the proposed classification method could appropriately classify the critical flow conditions in which large and unstable pressure fluctuations are observed. This method may be used to avoid critical operating conditions that lead to blade failure.