Abstract
The development of an enhanced neural network training algorithm and program for the calibration of velocity measurement instrumentation is presented. The backpropagation-based program, PROBENET, is intended to be a robust self learning code, developed to calibrate multi-port pressure probes for use in wind tunnel flow analysis. The code offers distinct advantages over commercial packages' in terms of maximum allowable network size, training convergence rates, flexibility in network architecture, and network optimization capabilities. PROBENET incorporates multiple activation functions per layer, as well as heuristics-based procedures for network architecture optimization. Techniques for local minima avoidance and convergence rate improvement, incorporated into the algorithm, include: momentum, variable learning rate, and batch mode processing. The study compares the performance of two types of network architectures: single activation function per layer and multiple activation per layer, and shows that the latter consistently produces a better solution in terms of convergence rate and accuracy. The, research contributes to the current techniques and methods for calibrating multi-port probes for flow measurement by implementing a robust neural network and training algorithm. The development of a novel nearly-onmi-directional 18-hole probe to overcome the flow angularity measurement limitations of traditional 5-and 7-hole probes is detailed and the application of neural network calibration to this new design demonstrates the flexibility and utility of the method. To prove the capabilities of PROBENET, a 5-hole hemispherical probe is used to demonstrate the accuracy of threecomponent velocity prediction over a large range of flow angularity. In addition, a unique mini-5-hole probe having a diameter of only 0.065" is calibrated and used to measure three dimensional velocity data about a delta wing model to demonstrate PROBENET's real world application.
Kinser Robert Eric (1996). Development of neural network calibration algorithms for multi-port pressure probes. Master's thesis, Texas A&M University. Available electronically from
https : / /hdl .handle .net /1969 .1 /ETD -TAMU -1996 -THESIS -K57.