# Projects in 1987–1988

## Aerojet Electrosystems: Outgassing and Contamination in Vacuum Systems

- Advisor
- Fiki Shillor Meir

Outgassing is a process by which molecules migrate out of a material body into its surroundings. This process can become the source of contamination, particularly in the near perfect vacuum conditions encountered in space application. The proposed project developed a predictive mathematical model of outgassing based on physical principles.

## General Dynamics: Multipath Modeling

- Advisor
- Janet Myhre
- College
- CGU

When the target object is close to the sea surface, radar signals reflect from that surface as well as from the object, and a signal may have traversed a number of paths on return to the emitter. This gives a number of ghost images. The latter might be reduced by choice of a frequency at which the sea surface is most absorbing. The goal of the Clinic was to identify that frequency. First a statistical model for the sea surface was constructed and its reflecting and absorbing properties analyzed. Then the radar power loss could be computed along the various paths and estimates made of the images.

## General Dynamics: Modeling and Simulation of Neural Network Image Classifiers

- Advisor
- Stavros Busenberg

This project investigated the feasibility of using neural networks to recognize two-dimensional objects with up to twenty degrees of rotational invariance. The investigation took two paths. The first was a continued mathematical analysis of general neural networks with the aim of determining the theoretical limits of rotationally invariant object recognition. The second was the refinement and systematic use of the ATHENA software to simulate a variety of neural networks which the theoretical work suggests as candidates for successful rotationally-invariant image classification. These two approaches complement each other and were pursued in parallel, with constant interaction between them.

## Jet Propulsion Laboratory: Optimal Data Collection for MOSFET Modeling

- Advisor
- Gunnar Andersson
- College
- CGU

Problem 1. Physical and geometrical parameters used in modeling the flow of current were not known and were hard to measure. They were inferred (“parameter extraction”) from current measurements. The clinic addressed how the latter can be reduced, retaining good accuracy on the extracted values.

Problem 2. Formulae were based on infinite length, infinite width models. Terms to correct for finite values could be added with coefficients identified by regression from measurements. The clinic addressed what values of length and width are measurements to be taken for optimal results.

## Jet Propulsion Laboratory: Benchmarking Time Warp Synchronization Mechanism

- Advisor
- David Fisher

JPL had developed an operating system based on the Time Warp synchronization mechanism, which is a method for extracting concurrency from asynchronous, object-oriented computer simulations. While initial speedup results had been obtained for a war game simulation running under Time Warp on the Caltech Mark II Hypercube, no other “benchmarking” of Time Warp had been completed.

## NASA Dryden Flight Research Center: Flutter Stability of Aircraft in Flight

- Advisor
- Henry Krieger

Flight flutter testing is conducted on new and structurally modified research vehicles to verify the absence of aeroelastic instabilities within the flight envelope. This testing is accomplished by establishing frequency and damping trends for all excited elastic modes of the structure. Frequency trends can be established with excellent accuracy. However, the estimation of damping is difficult and results in appreciable data scatter. Methods currently being used to estimate frequency and damping from response data include fast Fourier transform, recursive filter and extended Kalman filter techniques. New techniques that predict the flutter boundary without requiring the estimation of damping were researched. The technique chosen through this research effort was to be rapid, accurate and capable of predicting flutter boundaries from random, forced, and impulse response data. A computer program written in Fortran demonstrated the technique on either actual flight data or computer generated data.