COSC Simulations Page

Transportation, Optimization and Scheduling Theory

Joint work with USAF's Air Mobility Command (AMC) in modeling, simulation and optimization of DoD's large-scale air transport operations. Development and implementation of a general model and methodologies for analyzing and optimizing large-scale air transportation networks, including time-window constrained routing and scheduling for the USAF. Examplaes include mathematical modelling, applications of Semantic Control and optimization to an AMC's simulation program, Mobility Analysis Support System (MASS), which attempts to simulate Air Mobility Command's airlift operations. Current effort includes optimization of DEDS simulations as well as formulation/simulation of on-the-ground activities.
(Click here for more details on the airfield simulation BRACE.)

Joint work with Systems & Electronics Inc.- St. Louis, MO
Project title: Decision and Control Project
Urban traffic prediction and management; vehicle routing and automatic control. Macro-level (traffic prediction and management) as well as micro level (vehicle routing and control).
(Please click here for more details and graphic representation.)

Semantic Control for Game Theory

Joint work with Rockwell International - El Segundo, CA
Project: Development of a Flight & Fire Control System which was tested in Germany by Messerschmidt and adapted as the pilot's assistant for the Advanced Euro Fighter.
Our contributions include:

Applications of Semantic Control paradigm to reduce higher-order diferential games to lower-order ones.

Logic programming and time-sweep algorithms for mission planning.

A neural network module for the identification and prediction of tactical air combat maneuvers based on incomplete information.

A simulation package for "low observables" (minimum-detection path planning), providing autonomous aircraft navigation in a time-varying environment cluttered with stationary & moving hostile radar sites.

A computer package for situation assessment and risk prioritization in medium-range (16-50 miles) air combat.

Joint work with Electronics & Space Corp. - St. Louis
Project title: Layered Defense Project (LDP)

Embedded control and decision-aiding system for situation assessment, guidance and control of a vehicle engaged in evasive maneuvers against primary and secondary pursuers.

System Identification and Intelligent Control

Joint work with McDonnell Douglas and NASA-Ames Research Center
Project: Intel. Flight Control Adv. Concept Prog.; On-Line Neural Network System Identification

Real-time system identification parameter estimation and control of a damaged F-15 aircraft. The Center's projects are:

development and implementation of methods for on-line model learning and parameter estimation via dynamical neural networks; and
implementation of a real-time Riccati solver for optimal control of the aircraft.

Joint work with McDonnell Douglas for an ARPA "dual-use" program
Project title: Intelligent Flight Control for the Fly-By-Light Advanced Hardware System (FLASH).

Implementation of a real-time (< 20 msec.) Riccati solver for the optimal control of a damaged aircraft.

Theory and application of artificial neural networks to problems of estimation, system identification and control:

Multi-layerd perceptrons: aircombat maneuver prediction; handwritten character recognition; network flow prediction for the Mobility Analysis Support System;

Radial-basis function neural networks: road traffic flow prediction;

Neurons with local memory: proofs of stability and convergence; application to control of a Boeing 727 through windshear;

Dynamic neural networks: development of a simulator for system identification via dynamic neural networks, as well as disturbance rejection and control using dynamic neurons.

Recurrent high-order neural networks: development of RHONNI simulator for on-line system identification and estimation; application to an F-15 aircraft.

System identification for semiconductor crystal growth and tuning of multi-loop controller gains. Improved models and control methodologies for crystal growth technologies (consulting work with MEMC Electronic Materials Inc.).

Neural Network Augmented Antiskid System
Joint work with Boeing Corporation to develop a neural network augmented anti-skid control system for MD-90 and similar transport aircraft. The controller is based on robust control theory and on-line system identification capabilities of feedforward neural networks. Extensive simulation tests indicate that the resulting control system improves stopping efficiency and may yield significant extension in tire and brake service life compared to the systems currently in operation.

Detailed presentation. (Download MS-PowerPoint Presentation source file.)

Medical Informatics

Project: Diagnosis Prediction via an Artificial Neural Network Knowledge Base

The use of expert systems as means of predicting medical diagnoses and recommending successful treatments has been a highly active research field in the past few years. Development of a medical expert systems that use artificial neural networks as their knowledge bases appears to be a promising method for predicting diagnosis and possible treatment routine. The purpose of this project was to construct and train an artificial neural network to serve capable of serving to serve as a dynamic 'look-up table' that can accurately classify medical diagnoses based on patients' given symptoms. The network may in the future serve as a knowledge base for an expert system specializing in medical diagnosis, testing evaluation, treatment evaluation, and treatment effectiveness. The project serves as the first component of a much larger system that will assist physicians facilitate the reasonable ordering tests and treatments and minimize unnecessary laboratory routines while reducing operational costs.

The network correctly classified 965 out of 1292 cases (74.7%) in the training set and 418 out of 554 cases (75.5%) in the testing set. In some cases of classification, the network'snetworks prediction appears to be reasonable even though they differ from the physician's diagnoses. The notion of "reasonable" can be implied if the treatment, need for further diagnostic testing, clinical follow-up, and outcome are equivalent. For instance, muscle strain was sometimes incorrectly diagnosed as abdominal pain if abdomen was the primary body part. If we accept these "reasonable" diagnoses, network performance increases by 6.5%.

The complete paper.