Before a helicopter leaves the plant, it needs to be tuned so that its vibrations meet the required specifications. Helicopter track and balance is currently performed based on “sensitivity coefficients” which have been developed statistically after years of production experience. The fundamental problem with using these sensitivity coefficients, however, is that they do not account for the nonlinear coupling between modifications or their effect on high amplitude vibrations. In order to ensure the reliability of these sensitivity coefficients, only a limited number of modifications are simultaneously applied. As such, a number of flights are performed before the aircraft is tuned, resulting in increased production and maintenance cost. In this paper, the application of feedforward neural nets coupled with back-propagation training is demonstrated to learn the nonlinear effect of modifications, so that the appropriate set of modifications can be selected in fewer iterations (flights). The effectiveness of this system of neural nets for track and balance is currently being investigated at the Sikorsky production line.

Issue Section:
Research Papers
Topics:
Artificial neural networks, Flight, Vibration, Aircraft, Assembly lines, Feedforward control, Maintenance, Reliability
1.
Fahlman, S. E., and Lebiere, C., 1990, “The Cascaded-Correlation Learning Architecture,” Technical Report No. CMU-CS-90-100, School of Computer Science, Carnegie Mellon University.
2.
Hertz, J. A., Krogh, A. S., and Palmer, R. G., 1991, Introduction to the Theory of Neural Computation, Addison-Wesley, Redwood City, CA.
3.
Jordan, M. I., and Rumelhart, D. E., 1992, “Forward Models: Supervised Learning with a Distal Teacher,” Cognitive Science, in print.
4.
Kung, S. Y., and Hu, Y. H., 1991, “A Frobenius Approximation Reduction Method (Farm) for Determining Optimal Number of Hidden Units,” Proc. IJCNN, IEEE, Seattle, WA, July, pp. 163–168.
5.
Matsuoka
K.
,
1991
, “
An Approach to the Generalization Problem in the Backpropagation Method
,”
Systems and Computers in Japan
, Vol.
22
, No.
2
, pp.
897
–
905
.
6.
Rumelhart, D. E., Weigend, A. S., and Huberman, B. A., 1991, “Generalization by Weight-Elimination applied to Currency Exchange Rate Prediction,” Proc. IJCNN, IEEE, Seattle, WA, July, pp. 837–841.
7.
Sanger
T. D.
,
1991
, “
Optimal Hidden Units For NonLinear FeedForward Neural Networks
,”
Pattern Recognition and Artificial Intelligence
, Vol.
5
, No.
4
, pp.
545
–
561
.
8.
Utans, J., and Moody, J., 1991, “Selecting Neural Network Architectures via the Prediction Risk: Application to Corporate Bond Rating Prediction,” Proc. First International Conference on Artificial Intelligence Application on Wall Street.
9.
Wada, Y., and Kawato, M., 1991, “Estimation of Generalization Capability by Combination of New Information Criterion and Cross Validation,” Proc. IJCNN, IEEE, Seattle, WA, July, pp. 1–6.
10.
Whitley, D., and Karunanithi, N., 1991, “Generalization in Feed Forward Neural Networks,” Proc. IJCNN, IEEE, Seattle, WA, July, pp. 77–82.
This content is only available via PDF.
Copyright © 1995
 by The American Society of Mechanical Engineers
You do not currently have access to this content.