Active structural control by neural networks

The present paper develops some alternative schemes to apply back‐propagation neural networks in the area of active structural control. The structural systems referred to are plane hysteretic structures, the restoring forces of which are idealized by an endochronic model. The control algorithms herein developed call for two different neural networks in order to approximate the forward and backward dynamics of the system. The latter network is the actual controller whereas the former one is needed to determine the response of a virtual system upon which no control forces are applied. Results from successful numerical examples are proposed with the objective of computing a quantitative measure of the effectiveness of the system.