This paper presents the preliminary results of using artificial neural networks in the prediction of gas side convective heat transfer coefficients on a high pressure turbine blade. The artificial neural network approach which has three hidden layers was developed and trained by nine inputs and it generates one output. Input and output data were taken from an experimental research program performed at the von Karman Institute for Fluid Dynamics by Camci and Arts [5,6] and Camci [7]. Inlet total pressure, inlet total temperature, inlet turbulence intensity, inlet and exit Mach numbers, blade wall temperature, incidence angle, specific location of measurement and suction/pressure side specification of the blade were used as input parameters and calculated heat transfer coefficient around a rotor blade used as output. After the network is trained with experimental data, heat transfer coefficients are interpolated for similar experimental conditions and compared with both experimental measurements and CFD solutions. CFD analysis was carried out to validate the algorithm and to determine heat transfer coefficients for a closely related test case. Good agreement was obtained between CFD results and neural network predictions.

A numerical study, based on experimental work of Inanli et al. [1] is conducted to understand the heat transfer characteristics of film cooled test plates that represent the gas turbine combustor liner cooling system. Film cooling tests are conducted by six different slot geometries and they are scaled-up model of real combustor liner. Three different blowing ratios are applied to six different geometries and surface cooling effectiveness is determined for each test condition by measuring the surface temperature distribution. Effects of geometrical and flow parameters on cooling effectiveness are investigated. In this study, Conjugate Heat Transfer (CHT) simulations are performed with different turbulence models. Effect of the turbulent Prandtl Number is also investigated in terms of heat transfer distribution along the measurement surface. For this purpose, turbulent Prandtl number is calculated with a correlation as a function of local surface temperature gradient and its effect also compared with the constant turbulent Prandtl numbers. Good agreement is obtained with two-layered k–ϵ with modified Turbulent Prandtl number.

No abstract