1. Two neural network algorithms, which are capable of modeling response surfaces, have been selected to integrate with multiple linear regression to compute precision intervals. In the first approach, the Radial Basis Function Network (RBFN) is introduced to compute precision intervals. A brief description of Radial Basis Function Network and the integrated computation procedure of Confidence and Prediction intervals with the linear regression analysis are given in Appendix I. As an example, the application to modeling force data as a function of angles of attack for the TST Alpha Jct Model as shown in Figure I is included for data' acquired from Tunnel 16T at the Arnold Engineering and Development Center (AEDC)
2. The force coefficients' were taken at Mach Number 0.8 and Chord Reynolds Number 1.5 millions under transition-free configuration in the present application. The angle of attack ranges from -4 to IO degree. The results obtained from the RBFN and the original data are plotted in Figure 2. The comparison of RBFN results and tunnel data is within the accuracy of tunnel measurement.The results of 95% Confidence Interval Half Width on the response surface from Eq. (1-4)are also shown in Figure 2 as the error band by the Coef-HI (or -UPPER) and Coef-LO (-LOWER). The results are satisfactory as expected. In Figure 3, the 95% CIHW (Confidence Interval Half Width) is and Coef-Residual, which is defined as