SHUNT CONTROL ON CANTILEVER BEAM BY NEURAL NETWORK : OBJECTIVE FUNCTION

Resumo

Piezoelectric materials have been extensively studied in recent years for the development of
electromechanical harvesting devices. Usually connected to a structure, these kinds of materials convert
kinetic energy into electric energy, and your electronic parameters interact directly to the vibrations of
the system they are coupled on. Therefore, this work aims at comparing the use of genetic algorithms and
artificial neural network techniques in the implement of shunt control in a structural set of a cantilever
beam coupled to a piezoelectric layer in the piezo-beam configuration. For the architecture of the genetic
algorithm and the neural network, was used a software with finite element model implemented and
the comparisons were made analyzing the computational demand of the algorithms and their respective
responses when both of them were defined on the task of finding the best combination between the
parameters of resistance and inductance of the piezoelectric patch that result in the best damping to the
structure. The comparison between the techniques had a focus on the use of the objective function of
the system by them, parameter used as a metric to gauge the aggregate computational demand, and the
damping provided with the respective configurations suggested by the two techniques. The results show
that the neural network after training completes your execution in order of 102 seconds, much faster
than the genetic algorithm, presenting a response with an average gain in damping of 23,24dB, but, even
though faster, this technique demands much more iterations than the genetic algorithm, due to its nature
of parallel computations, and additional care for the input data, that need a pre-processing not seen in the
genetic algorithm technique.