Prediction of blade life cycle for an industrial gas turbine at off-design conditions by applying thermodynamics, turbo-machinery and artificial neural network models

A novel method for estimating the rotor blade life cycle of an industrial gas turbine (GT) by the use of artificial Neural Network is proposed in this paper At the first step the blade life cycle is obtained by the use of Larson-Miller method which uses output results of GT performance modeling and blade thermal-mechanical data. Then results of rotor blade life cycle analysis by the above method are compared with results of stress factor curve (which is provided by manufacturers). Comparison of results revealed an average difference value of 9.7 % between blade life cycle estimation by two above mentioned methods. In the next step, by input data such as mass flow rate, temperature and pressure of hot flue gas, the output data such as blade cooling air and turbine shaft rotational speed are obtained from GT modeling. Then blade life cycle are also obtained by Larson-Miller​ method for 811 sample points of GT operating conditions for various ambient temperatures and load ratios. These data are used for neural network training. Results show that life cycle estimated values by neural network method in comparison with life cycle estimated values by Larson-Miller method, had about 4.8% error value in maximum (with 10-4 as mean square error, MSE). Finally, by the use of neural network method, the effects of gas turbine operating and health conditions (at various ambient temperatures, GT load ratios and compressor fouling levels) on blade life cycle are investigated. If we expect to get the nominal power output of clean blade at ISO ambient condition, in ambient temperature range of 15 to 45 ºC, the GG turbine first rotor blade life cycle reduces from 4.85 to 0.07 and in the range of 0 to 7% compressor fouling, turbine blade life cycle reduces from 4.85 to 0.68 years.