Improving large angle flow measurement accuracy of five-hole probe using novel calibration coefficient and accuracy progressive neural network

Abstract

Novel calibration coefficients and data-fitting techniques based on a two-stage accuracy progressive neural network were developed to improve the accuracy of a five-hole probe in measuring large-angle flows. By modifying the denominator and numerator of traditional calibration coefficients, the novel coefficients can solve the singularity and multi-value problems of large-angle flow measurement. By training the neural network using both global calibration data and the calibration data of large measurement error points, the two-stage accuracy progressive neural network can effectively improve the measurement accuracy of a five-hole probe when flow separation occurs around the probe head at large flow angles. The experimental results demonstrate that applying the novel calibration coefficients and accuracy progressive neural network ensures that the calibration error of the flow angle is less than 0.8°, and the flow pressure error is less than 0.1 % when the flow angle reaches 50° at low subsonic speeds.