Active vibration control of structural systems with a preview of a future seismic waveform generated by remote waveform observation data and an artificial intelligence–based waveform estimation system

Abstract

We propose a new active vibration control strategy of structural systems based on the information of future seismic waveform observed in remote observation sites. The observed waveform information of the remote site is transmitted by a waveform transmission network to the structure under control. The waveform transmission network is realized by interconnecting multiple controlled structures and observation sites. By using the remote waveform containing the future information of the disturbance at the location of the controlled structure, we propose an active control method that achieves fairly higher control performance over conventional methodologies. A preview control consisting of the state-feedback and feedforward control (preview action) is adopted as the control law. For the preview action, a future seismic waveform in some time interval is needed. Because the future seismic waveform is not available, the preview action contributing the performance improvement is generally impossible. To get over this difficulty, an artificial intelligence–based waveform estimation system to estimate the future seismic waveform is proposed. The core of the wave estimation system is a multi-layered artificial neural network. Through a small-scale simulation study with a recorded seismic event in Japan, we show that the proposed control method achieves much higher control performance over the optimized [Formula: see text] state-feedback control law.

Generalized Performance of Neural Network Controllers for Feedforward Active Control of Nonlinear Systems

“…NNs in particular are known to possess the property of being 'universal approximators' [9] and are therefore an attractive blackbox method for the modelling and control of unknown or uncertain nonlinear systems. Many different uses of NNs have been studied, including plant/system modelling [4,[10][11][12][13], feedforward controller design [4,10,14], inverse modelling [15], signal prediction and feedback control [16][17][18][19][20], linear filter selection [21], adaptive parameter estimation for linear controllers [20,22], frequencydomain control [23], multichannel controller design [24], and signal classification [25]. The similarity in structure between NNs and linear filters provides good motivation DOI: 10.61782/fa.2023.0050 3599 for their use in both system modelling and feedforward controller design.…”

Active seismic control using neural network

“…Munteanu et al [6] have investigated the active control technology and their benefits in the case of seismic structural behavior. Hiramoto et al [7] have developed an active vibration control strategy consisting of state-feedback and feedforward control and used multi-layer artificial neural network for generating future wave form of seismic disturbance. Chang et al [8] presented a neuro-control system that generates the best control signal to lessen vibration during earthquakes is based upon structural modal energy.…”

Seismic Response Reduction Using an Active Tuned Mass Damper Driven by an Artifical Neural Network Controller

“…The artificial neural network was brought to structural engineering back in the 1980's Adeli and Yeh (1989) [6]. It has gained a lot of interest in designing active control schemes in vibration control [7][8][9][10]; due to their parallelism and learning capabilities, they can produce good performance with a limited number of sensors. The self-learning capability of artificial neural network control above traditional control is a benefit since it eliminates the requirement for the previous system's information, where it must adapt to environmental changes.…”