An improved linear quadratic regulator control method through convolutional neural network–based vibration identification

Abstract

An efficient vibration control can reduce negative effects induced by environmental vibrations and thereby improve the performance of precision instruments and the qualities of manufacture. The performance of the widely used linear quadratic regulator control algorithm, a classical active control methodology, depends on the parameters of the control algorithm. Consequently, a set of fixed parameters cannot satisfy the demand for controlling various types of environmental vibrations. Therefore, this study proposes a vibration identification method based on a convolutional neural network. This method helps to optimize the linear quadratic regulator algorithm by selecting corresponding optimal parameters according to the identification results, thereby achieving the objective of optimal control subjected to various types of vibration inputs. Specifically, environmental vibration signals are collected, and the preliminary features of the vibrations (i.e. wavelet coefficient matrices or images) are adopted as input samples for the convolutional neural network. A genetic algorithm is used to optimize the parameters of the linear quadratic regulator algorithm for each type of vibration; subsequently, the trained convolutional neural network model with the best performance is used to identify the vibration and select the corresponding optimal parameters of the linear quadratic regulator algorithm under different types of vibration inputs. Case studies show that the performance of the improved linear quadratic regulator control method is significantly better than that of the conventional linear quadratic regulator algorithm with fixed parameters.

Uncertainty Analysis and Experimental Validation of Identifying the Governing Equation of an Oscillator Using Sparse Regression

“…Due to the advantages of LSPL over LSST mentioned in Section 3, LSPL was selected to experimentally identify the oscillator using the signals from the measurement. The identified models were then compared with the theoretical linear model in Equation (19) to evaluate the performance of this identification method.…”

“…These methods are especially useful when the first principle knowledge is not discovered, for example, the macroscopic behavior of advanced materials [7], in aerospace engineering [8], in biology [2,9,10], in economy [11], in music [12,13], etc. The models of vibration systems can be identified with various approaches, such as iterative model updating [14], the Polynomial NonLinear State Space (PNLSS) model [15], Volterra series [16], the Wiener and Hammerstein model [17,18], artificial neural networks [19,20], and equation-free model building [21]. On the downside, these methods usually build "black-box" or "grey-box" models, which means the theoretical understanding of the system is totally or partially absent from the model [4].…”

A deep learning approach to rapid regional post‐event seismic damage assessment using time‐frequency distributions of ground motions

“…Kim et al 31 implemented a CNN model to extract the features of structural hysteretic behavior. There are also some studies which use CNN in the field of structural health monitoring or seismic response modeling 32–35 . Xu et al 36,37 uses other kinds of input for near real‐time seismic damage prediction, including the seismic IMs and acceleration records.…”

A Monte Carlo Simulation Approach in Non-linear Structural Dynamics Using Convolutional Neural Networks

“…With special emphasis on problems in earthquake engineering, Sun et al (2021) has summarized the incorporation of machine learning approaches to response, damage, and failure prediction. Convolutional neural networks and deep learning have been used to predict the whole response time history of the structure subjected to a transient excitation (Zhang et al, 2020) and wavelet transformation-based response prediction (Lu et al, 2020(Lu et al, , 2021Liao et al, 2021). Thaler et al (2021a,b) proposed a machine leaning-enhanced Monte Carlo method using a simple feed-forward neural network.…”