Modeling of piezoelectric actuator based on genetic neural network

Piezoelectric actuator is widely used in precision positioning mechanism for the advantages of ultra high resolution, high response frequency and rapid dynamic performance. But the displacement error is conducted for the inherent hysteretic nonlinear characteristics, and the tracking precision is limited. A modified modeling method combining the neural network with the genetic algorithm (GA) is designed in this paper to improve the modeling performance. The mechanical structure is analyzed, and a Bouc-Wen model is introduced to express the nonlinear kinetics. A three-layer neural network is applied to identify the parameters including the weight and threshold values by Levenberg-Marquardt algorithm. GA is used to achieve the optimized solution of the network parameters. The data pairs including actuating voltage and corresponding displacement are regarded as the samples to train the network off-line. A low frequency triangle voltage with variable amplitude is applied to validate the effectiveness of the proposed method. The results show that the mean positioning error is reduced from 0.39μm to 0.24μm, and the maximum error from 0.76μm to 0.33μm respectively compared with the static neural network. A more accurate model is established for the control system design in the future.