Direct inverse control for active vibration suppression using artificial neural networks

Abstract

In this article, a control method based on artificial neural networks applied to the vibration control of flexible structures is presented. The direct inverse control method is used. This method consists in the identification of the inverse dynamics of the plant using an artificial neural network to be used as the controller. An application example is proposed, and two problem variations are treated. The application problem is based on a cantilever plate model. The plate model is obtained using the finite element method. For the first problem, the controller is designed using the full-order plant model. In the second example, a model reduction is made to evaluate the performance of this technique applied to control problems with dynamic uncertainties. The results were evaluated according to the time response and frequency response of the closed-loop system. To compare the results obtained using the control method based on artificial neural networks, the previous examples were also solved using the [Formula: see text] control method. The obtained results show that the control method based on the inverse model using neural networks is effective in solving this kind of problem.

A review of machine learning methods applied to structural dynamics and vibroacoustic

“…De Abreu et al [190] implemented a direct inverse NN control of a vibratory system by training an NN as the inverse model of the plant, such that the NN receives the current state and the desired state and outputs the actuator signal. Similarly, Ariza-Zambrano and Serpa [191] applied direct inverse NN control to a beam cantilever, in which the NN was trained both with a full-state FEM model and with a ROM to account for dynamic uncertainties in practical scenarios, showing more stable results than Hinfinity control. Nerves and Krishnan [192] used NN direct controller to control wind-induced vibrations in a building-TMD (tuned mass damper) system by considering the plant as the output layer of the NN, as in a feedback linearization control.…”

A Review of Machine Learning Methods Applied to Structural Dynamics and Vibroacoustic

“…De Abreu et al [112] implemented a direct inverse NN control of a vibratory system by training an NN as the inverse model of the plant, such that the NN receives the current state and the desired state and outputs the actuator signal. Similarly, Ariza-Zambrano and Serpa [113] applied direct inverse NN control to a beam cantilever, in which the NN was trained both with a full state FEM model and with a reduced model to account for dynamic uncertainties in practical scenarios, showing more stable results than H-infinity control. Nerves and Krishnan [114] uses NN direct controller to control wind-induced vibrations in a building-TMD system, by considering the plant as the output layer of the NN, as in a feedback linearization control.…”

Seismic response of high-rise buildings using long short-term memory intelligent decentralized control system

“…The neural network control algorithm is inspired by the organization of neurons in the brain. The neural network structure is formed by mathematical modelling, which is especially suitable for modelling and control of complex systems with large uncertainties, nonlinearities, and time-delaying (Ariza-Zambrano and Serpa, 2021; Bozorgvar and Zahrai, 2019; Wang et al, 2020b). Neural networks have evolved from the earliest neurons (Brown and Yang, 2001), BP neural networks (Gao and Liu, 2016), RBF neural networks (Nian et al, 2020; Sun et al, 2018), and other shallow learning neural networks to the current deep-learning network (Schmidhuber, 2015) in the control field.…”