Local heat transfer distribution on smooth and roughened surfaces under an array of angled impinging jets

Measurements of the local heat transfer distribution on smooth and roughened surfaces under an array of angled impinging jets are presented and used to validate a computational model for predicting impingement heat transfer coefficients. The test rig is designed to simulate impingement with cross-flow in one direction which is a common method for cooling gas turbine components such as the combustion liner. The jet angle is varied between 30, 60, and 90 degrees as measured from the impingement surface, which is either smooth or randomly roughened. Liquid crystal video thermography is used to capture surface temperature data at 5 different jet Reynolds numbers ranging between 10000 and 35000. The effects of jet angle, Reynolds number, gap, and surface roughness on heat transfer efficiency and pressure loss are determined along with the various interactions among these parameters.; Experimental results show that peak heat transfer coefficients are highest for orthogonal jets impinging on a roughened surface. The ratio of peak to average Nu is lowest for 30-degree jets impinging on roughened surfaces. It is often desirable to minimize this ratio in order to decrease thermal gradients, which could lead to thermal fatigue. Peak heat transfer coefficients decay in the cross-flow direction by close to 24% over a dimensionless length of 20. The decay of Nu is lowest for the case of 30-degree jets impinging on a roughened surface where the decay was less than 3%. The decay is greatest for 30-degree jet impingement on a smooth surface where the decay exceeded 23% for some Reynolds numbers.; Numerical models of the jet plate impingement configurations for the cases of smooth surfaces are developed to predict surface Nusselt number distributions. Various turbulence models are examined including the standard k-epsilon model and several low Reynolds number turbulence models. Numerical results for the Nusselt number distribution are obtained using the standard k-epsilon model and the Yang-Shih low Re model for all cases and compared with test results. The Yang-Shih model was able to predict average Nusselt number within 2 to 30%. The standard k-epsilon model predicts average Nu within 0 to nearly 60% error.