Accession Number : ADA323742

Title :   A Bayesian Classifier Based on a Deterministic Annealing Neural Network for Aircraft Fault Classification.

Descriptive Note : Final technical paper, Mar-Dec 95,

Corporate Author : ARMSTRONG LAB WRIGHT-PATTERSON AFB OH HUMAN RESOURCES DIRECTORATE

Personal Author(s) : Wang, Jun ; Chu, Shing P.

PDF Url : ADA323742

Report Date : JAN 1997

Pagination or Media Count : 15

Abstract : A Bayesian classifier based on a recurrent neural network was developed for aircraft fault classification. From historical maintenance data the posterior probabilities of fault classification based on given fault indicators are estimated and derived using the Bayes' rule. Based on Bayesian decision theory, the fault classification problem is formulated as a linear integer programming problem to minimize an expected loss function using the posterior probabilities. The linear integer programming problem is then converted equivalently to a standard linear programming problem. A two layer recurrent neural network is used to carry out the computation task for fault classification by solving the formulated linear programming problem. The simulation results of a pilot study based on the synthetic data on the fire control radar system in F-16 aircraft show that the neural network approach is capable of real-time aircraft fault classification.

Descriptors :   *NEURAL NETS, *AIRCRAFT MAINTENANCE, *BAYES THEOREM, *COMPUTER AIDED DIAGNOSIS, COMPUTERIZED SIMULATION, REAL TIME, LINEAR PROGRAMMING, INTEGER PROGRAMMING, JET FIGHTERS, DETERMINANTS(MATHEMATICS), FIRE CONTROL RADAR, AIRCRAFT FIRE CONTROL SYSTEMS, TROUBLESHOOTING, STATISTICAL DECISION THEORY.

Subject Categories : Cybernetics
      Attack and Fighter Aircraft

Distribution Statement : APPROVED FOR PUBLIC RELEASE