Neural Network Based Auxiliary Controller for Online Signal Correction of Electric Linear Motor Shake Table
Selma H. Larbi1, Nouredine Bourahla2, Hacine Benchoubane3, Mohamed Badaoui4
1 Graduate student, Civil Engineering Dept. University of Blida1, Algeria
selma.h.larbi@gmail.com
2 Professor, Civil Engineering Dept. University of Blida1, Algeria
nbourahla@univ-blida.dz
3 Professor, Aeronautic Dept. University of Blida1, Algeria
h_benchoubane@yahoo.com
4 Assistant Professor, Civil Engineering Dept. University of Djelfa, Algeria
badaoui.mohamed@yahoo.fr
Abstract. High accuracy and precision in reproducing seismic signals on shaking tables is more and more needed for various applications in structural dynamic testing. However, the distortion between the command signal and the reproduced signal caused by inherent nonlinearities in the system and the nonlinear interaction between the table and the specimen is still defying the tremendous progress in control systems. In this paper, a Feedforward Neural Network is proposed to provide an online auxiliary correction to the command signal in order to drive the table to the desired acceleration trajectory. To investigate the potential of this strategy, a realistic model of a linear motor shaking table controlled by a PID-feedforward system (Quanser shake table III) is developed in Matlab/Simulink environment on the basis of recorded experimental data. The validated model was used then to generate a database by applying several real earthquake records. The obtained inputs and outputs were used to train, test and validate the Neural Network. The latter is integrated into the shaking table controller in an online mode to compensate the errors simultaneously with the PID and the Feedforward command signals. The response of the entire system has been enhanced with the additional online NN controller. The reproduced signals matched better the desired ones and the coefficient of correlation R between the output system and the target for several signals yield higher values. This enhancement is also felt in the frequency steady state behavior of the shaking table system.
Keywords: Shaking Table Control, Neural Network, Feedforward Neural Network, PID.
