Neural Networks Used to Compare Designed and Measured Time-Average Patterns

Electronic time-average holograms are convenient for comparing the measured vibration modes of fan blades with those calculated by finite-element models. At the NASA Lewis Research Center, neural networks recently were trained to perform what had been a simple visual comparison of the predictions of the design models with the measurements. Finite-element models were used to train neural networks to recognize damage and strain information encoded in subtle changes in the time-average patterns of cantilevers. But the design-grade finite element models were unable to train the neural networks to detect damage in complex blade shapes. The design-model-generated patterns simply did not agree well enough with the measured patterns. Instead, hybrid-training records, with measured time-average patterns as the input and model-generated strain information as the output, were used to effect successful training. One inspection process is outlined in the figure.

A twisted blade appears at the top left. The full time-average or characteristic pattern of the first vibration mode is shown next. The third and fourth pictures at the top show measured-region time-average patterns for undamaged and cracked blades, where a crack was induced by high-cycle fatigue. These patterns were sampled on a nonuniform finite-element-model grid (not shown). The neural networks processed the samples as often as 30 times/sec. The outputs of the neural networks, in this case, were chordwise strains.

Three kinds of neural-net training have been implemented with software. These are listed in increasing order of effectiveness.

1. Neural networks can be model trained with model-generated time-average patterns and model-generated strain patterns. The effectiveness of this technique depends strongly on the accuracy of the models.

2. Neural networks can be trained with measured time-average patterns and model-generated strain patterns.

3. Neural networks can be trained very effectively to categorize measured time-average patterns. Categories can consist of damaged and undamaged fan blades, for example.

We plan to expand this work in the future from nonrotating to rotating fan blades.

Bibliography

Decker, A.J., et al.: Comparison of Computational-Model and Experimental-Example Trained Neural Networks for Processing Speckled Fringe Patterns. NASA TM-1998-208814 (also International Conference on Optical Technology and Image Processing in Fluid, Thermal, and Combustion Flow, Proceedings of VSJ-SPIE98, Yokohama, Japan, Dec. 6-10, 1998, CD ROM format only), 1998.

Lewis contact: Dr. Arthur J. Decker, (216) 433-3639, Arthur.J.Decker@grc.nasa.gov
Authors: Dr. Arthur J. Decker, E. Brian Fite, Oral Mehmed, and Scott A. Thorp
Headquarters program office: OAT, DDF
Program/Project: AST

Next article

Previous article

[Table of Contents] [Index of Authors and Contacts] [Glenn Research & Technology Reports for Other Years] About the name change from Lewis to Glenn] [Word Search of Glenn's WWW sites]

Last updated June 16, 1999, by Nancy.L.Obryan@nasa.gov

Responsible NASA Official: Gynelle.C.Steele@nasa.gov
216-433-8258

Point of contact for NASA Glenn's Research & Technology reports: Cynthia.L.Dreibelbis@nasa.gov
216-433-2912
SGT, Inc.

Web page curator: Nancy.L.Obryan@nasa.gov
216-433-5793
Wyle Information Systems, LLC

NASA Web Privacy Policy and Important Notices