Fault detection of industrial large-scale gas turbine for fuel distribution characteristics in start-up procedure using artificial neural network method

Abstract

In this study, artificial neural network (ANN) based on operating data is adjusted to detect fault operation of large-scale industrial gas turbine. To detect faults, the operating characteristics of gas turbine are first predicted for fuel injection characteristics because, in general, lots of the operating failures are occurred in start-up procedure when fuel distribution characteristics are frequently changed. Data from the GE 7FA gas turbine were used, and the fuel distribution and other operating parameters are set by input and output parameters, respectively, to analysis effect of fuel distribution. The start-up procedure is divided with eight operating process (OP) to detect fault operation by fuel distribution characteristics. An overheating start is detected when turbine exhaust temperature (TET) reaches 110%–120% of the design value with the conventional detecting method; by contrast, the proposed method can detect this problem in advance before OP4. Further, sensitivity analysis of gas turbine operating characteristic parameters for each nozzle was performed to detect not only operation failures but also problems in the fuel supply system. Fault detectability was different with each OP and nozzle, and nozzle system abnormalities can be detected in advance by the results of sensitivity analysis.