Share Email Print

Proceedings Paper

Passive vibration tuning with neural networks
Author(s): Eric D. DiDomenico
Format Member Price Non-Member Price
PDF $14.40 $18.00

Paper Abstract

Neural network control of flexible structures demonstrates better settling time and energy dissipation than linear design methods. Optimal tuning of passive vibration absorbers for reduced order control is examined using linear and non-linear cases. Quasi-Newton (BFGS) and simplex optimization methods improved the Den Hartog parameters where unsupervised LMS or backpropagation techniques were unstable. Lessons on unsupervised training for dynamic system control are illustrated by examining convergence to the solution in `error space' (parameters vs. cost). Spring stiffness and passive damping of a reaction mass actuator (RMA) are `tuned' for best disturbance rejection using total energy as a cost function. A single neuron using two weights (one for damping and the other for the spring coefficient) improved beam energy over the Den Hartog parameters for the linear bi-modal case. The non-linear case demonstrates even better performance. A multiple layer network is then demonstrated for both the linear and non-linear cases. Optimization techniques improved linear system parameters when initiated at the linear solutions. Lab data for the linear single neuron case validates model fidelity.

Paper Details

Date Published: 1 May 1994
PDF: 11 pages
Proc. SPIE 2193, Smart Structures and Materials 1994: Passive Damping, (1 May 1994); doi: 10.1117/12.174093
Show Author Affiliations
Eric D. DiDomenico, U.S. Air Force Academy (United States)

Published in SPIE Proceedings Vol. 2193:
Smart Structures and Materials 1994: Passive Damping
Conor D. Johnson, Editor(s)

© SPIE. Terms of Use
Back to Top