Neural network controller for nanopositioning of a smooth impact drive mechanism

Abstract

In this paper, neural network theory is used to improve the positioning accuracy of smooth impact drivemechanisms (SIDMs), by designing a displacement controller that consists of a neural network identification (NNI) anda neural network controller (NNC). The dynamics of the SIDM are described by the NNI, which consists of an inputlayer, hidden layer, and output layer. The parameters of the NNI are adjusted using back propagation. The NNC isdesigned as a proportional-derivative (PD) controller, which is used to accurately control the displacement of the SIDM.The PD parameters are adjusted with an adaptive adjustment algorithm. A prototype of the SIDM was fabricated andan experimental control system was built that consists of a laser displacement sensor, power amplifier, data acquisitionboard, and SIDM prototype. The experimental results show that nanoscale positioning accuracy can be obtained. Thecontrol system can maintain steady operation, even if the output load mass is changed.