Date of Award

Spring 5-2007

Document Type

Thesis - Open Access

Degree Name

Master of Science in Aerospace Engineering

Department

Aerospace Engineering

Committee Chair

Eric v. K. Hill

Committee Member

Frank J. Radosta

Committee Member

Seth-Andrew T. Dion

Abstract

Through the use of an acoustic emission (AE) data acquisition system, a Kohonen self-organizing map, and a back-propagation neural network, AE data from 7075-T6 aluminum specimens were used to classify failure mechanisms and predict the number of fatigue cycles to failure. AE waveforms were captured from 40 notched tensile specimens during the low-cycle fatiguing process. A Kohonen self-organizing map and initial data filters were used to classify the data into two distinct failure mechanisms, plane strain and plane stress fracture, plus a third less prevalent mechanism. These results were employed to construct a back-propagation neural network to predict the number of cycles to failure from the first 250 cycles of AE data. Due to a scarcity of AE data, optimal prediction results were not obtained on all 40 specimens. However, a smaller set of 18 specimens, 9 for training and 9 for testing, produced a worst case prediction error of-13.9%.

Share

COinS