Date of Award

Summer 1992

Document Type

Thesis - Restricted

Degree Name

Master of Science (MS)

Department

Electrical and Computer Engineering

First Advisor

Feng, Xin

Second Advisor

Brown, Ronald H.

Third Advisor

Belfore, Lee A.

Abstract

Artificial Neural Network and Time Series Analysis are two emerging technologies. Neural Networks provide the potential for massive parallel computing architectures that have many computational advantages over traditional algorithm based methods. Studies of time series have shown that the Dynamic Data System method is a powerful tool for analyzing time series from a systematic point of view. This research is concerned with performing noisy dynamic signal identification by using a hybrid time series and neural network method. Simulated vibration signals are generated and time series anallysis methods are applied to estimate dynamic models for the observations, then the frequency decompositions of the signals are obtained from their estimated parameters. By feeding these sequences to back propagation neural networks, the frequency components are identified. The results demonstrate that the combined time series and neural network method is an effective tool for vibration signal identification in noisy environments. It also can be extended to other dynamic signal identification/classification applications. The hybrid time series and neural network method has the potential for building a connectionist (or neural network) expert system to identify noisy dynamic signals. This implementation has several advantages over traditional expert systems and is suitable for real world applications.

Share

COinS

Restricted Access Item

Having trouble?