标题：Long Short-term Memory Neural Network-based System Identification and Augmented Predictive Control of Piezoelectric Actuators for Precise Trajectory Tracking
摘要：AbstractHigh-speed and high-precision systems based on piezoelectric actuator (PEA) demand precise real-time trajectory tracking. Model-based control techniques have been proven effective in achieving desired tracking accuracy. However, modeling uncertainty and linearization losses are the biggest hurdles in these techniques to achieve high precision performance over broad frequency ranges at high speed. To overcome these limitations, in this work, we propose a long short-term memory (LSTM) neural network-based inverse system identification and augmented predictive control using a linear model predictive control (MPC) to achieve high precision trajectory tracking of PEAs. An LSTM network was built and trained to model the inversion dynamics of the PEA system. The benefit of using LSTM is that it ensures the long-term dependencies of time series data, and hence it can model the system dynamics for both low and high-frequency ranges. Once the LSTM inversion model accuracy was evaluated, it cascaded with the commercial PEA, which together was mostly linear. This combined system was controlled by augmenting it with a linear MPC controller. The use of augmented predictive control using the nonlinear LSTM inversion model led to improved modeling accuracy and higher speed of operation with a reduced computational load. The results demonstrated the efficacy of the proposed approach in real-time high-speed trajectory tracking of PEAs.
关键词：KeywordsNeural networksNonlinear system identificationIdentification for controlModel predictive controlTrackingReal-time controlPiezoelectric actuators
