Neural-Network-Based Constrained Output-Feedback Control for MEMS Gyroscopes Considering Scarce Transmission Bandwidth

• Published in: IEEE transactions on cybernetics Vol. 52; no. 11; pp. 12351 - 12363
• Main Authors: Shao, Xingling, Shi, Yi
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.11.2022; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Artificial neural networks; Bandwidth; Bandwidths; Control systems; Control theory; Disturbances; Feedback control; Gyroscopes; Hybrid quantizer (HQ); learning complexity; microelectromechanical system (MEMS) gyroscopes; Microelectromechanical systems; Micromechanical devices; neural adaptive; Neural networks; Nonlinear systems; Output feedback; Quantization (signal); Stability analysis; State variable; Velocity measurement
• ISSN: 2168-2267; 2168-2275; 2168-2275
• DOI: 10.1109/TCYB.2021.3070137

In this article, a neural-network-based constrained output-feedback control is considered for microelectromechanical system (MEMS) gyroscopes subject to scarce transmission bandwidth and lumped disturbances resulting from model uncertainties, dynamic coupling, and environmental disturbances. First, a hybrid quantizer capable of achieving an adjustable communication rate and quantization density is proposed to convert continuous control signals into discrete values, allowing for reduced chattering behavior even when control actions vary within large regions and enhanced tracking accuracy can be ensured. Subsequently, by applying two types of nonlinear mapping, all state variables of MEMS gyroscopes are restrained within the predefined time-varying asymmetric functions without imposing stringent feasibility conditions on virtual control laws. Furthermore, an echo-state network-based minimal learning parameter neural observer is developed to simultaneously recover the unmeasurable velocity-state variables, matched as well as unmatched disturbances in constrained MEMS gyroscopes dynamics, enabling an output-feedback control solution with a decreased online learning complexity. It is shown via the Lyapunov stability and nonsmooth analysis that all signals in the closed-loop system remain ultimately uniformly bounded even with discontinuous control actions. Comparison simulations are produced to certify the effectiveness of the presented controller.