Abstract
A robust and accurate neural network based algorithm phics. for the calibration of miniature multi-hole pressure probes has been developed and a detailed description of its features and use is presented. The code that was developed was intended to handle both the a-hole and 7-hole configurations of the pressure probes in incompressible as well as the compressible flow regimes. The code was also intended to overcome the deficiencies like network size, convergence rates, flexibility in network architectures and network optimization abilities in the commercially available packages (Demut and Beale's Neural Network Toolkit). The code also incorporates multiple activation functions per layer that is not possible in commercial codes. It also has built into it several techniques that would accelerate convergence namely, momentum learning, variable learning rate and batch mode processing. The current research contributes to the techniques and methods for calibrating multi-hole pressure probes for flow measurement by implementing a robust neural network-training algorithm.
Vijayagopal, Rajesh (1998). Neural network calibration for miniature multi-hole pressure probes. Master's thesis, Texas A&M University. Available electronically from
https : / /hdl .handle .net /1969 .1 /ETD -TAMU -1998 -THESIS -V55.