Control of linear smart structural systems using neural networks

In recent studies, neural network based controllers for vibration suppression of smart structures have been reported. Many of these controller have been successfully implemented in simulation as well as using PC based data acquisition hardware. These studies have shown that in addition to conventional controller design methodologies, neural networks offer an effective basis for design and implementation of controllers. With the introduction of the Electronically Trainable Analog Neural Network (ETANN) chip i80170NX by Intel and a digital neural network chip Ni1000 by Nestor Corp., hardware implementation of neural network based controllers has been made possible. These neural network chips have also found applications in other areas such as signal processing and character recognition. In this paper, the capabilities of the ETANN based robust controllers for smart structural systems have been investigated. Robust controllers like the Liner Quadratic Regulator (LQR) and Linear Quadratic Gaussian with Loop Transfer Recovery (LQG/LTR) are implemented on a cantilevered plate system using the ETANN chip. Specially shaped PVDF film is used as sensors and PZTs as actuators. The LQG/LTR controller is implemented in two neural network configurations for dynamical systems suggested by Narendra and Parathasarathy. Analog hardware components used in the interface between the ETANN chip and the actuators/sensors on the smart structure test article have been developed. Practical considerations and limitations of the fully analog implementation of the controllers which are not considered in simulations have been discussed in the paper. Practical consideration in training the analog neural network chip for optimal performance has also been described. Experimental results of the closed loop performance of the smart structural system are presented.