Keywords: wind, prediction, control, mass-damper, tall building, inverse model, neural network.

Recently, significant researches have been conducted in developing and verifying innovative active control algorithms for earthquake-excited buildings. Similarly, several control devices and sensors have been proposed and tested. In this paper, a neural network based algorithm is proposed and evaluated for a wind-excited tall building equipped with active tuned mass damper (ATMD). The used building model is a benchmark model proposed in the second world conference on structural control [1]. A neural network model replicates and predicts the structure response due to wind excitations is trained and evaluated as part of the controller design. A second inverse-dynamics model is prepared using neural network to complete the controller setup. The structure evaluation model obtained from a benchmark problem. The trained neural network response-predictor as well as the inverse-dynamics neural network are being employed to control the building online. The algorithm of the controller design and the process of controlling the structure are presented herein.

Artificial neural networks are massively parallel computational models for knowledge representation and information processing. Neural network have some unique humanlike capabilities in information processing. Probably, learning from examples is the most important capability in information processing. Neural networks are capable of learning complex highly nonlinear relations and associations from a large body of data due to their intrinsic nonlinearity, adaptability, noise immunity, generalization ability and robustness.

Most recent efforts in wind-excited structures were designed as static forces although wind is a turbulent flow, characterized by random fluctuations of velocity and pressure. Real proper wind dynamic investigations initiated with the failure of the Tacoma Narrows Bridge in 1940. Yang and his co-authors [2] developed a series of structural control benchmark problems for wind excitations and they modelled the wind loads by: Stochastic processes; and a set of sample functions of wind forces simulated from the wind spectra. A benchmark problem for the response control of wind-excited tall buildings was proposed for control methods verifications. The building considered is a 76-story 306 meters concrete office tower. Hence, in this paper the proposed neuro-control method is verified using the benchmark problem.

The controller was developed and trained to control wind-excited 76-story building. The objective of this controller was to minimize the response of the structure. The controller, (referred to as NNPI) was assemblage of two neural networks the first one was trained to predict the future time response (referred to as PNN) and the second one was trained to learn the inverse-dynamics model (referred to as INN). Both neural networks are arranged online and operate jointly to produce the control force; hence the first controller (NNPI) is in operation. This controller showed significant capability in learning the control function. This controller showed robust performance and effective control force as well as simplicity in application. The controller is investigated and its effectiveness is demonstrated.

In this paper it is shown that neural networks are promising and effective alternatives for conventional structural control algorithms. Additionally, controllers based neural networks are adaptive, simple, nonlinear models that enjoy stability and robustness. The back-propagation neural networks strategy and the consecutive training stages are presented. Moreover, the plant predictor neural network is used to predict the structural response for short time periods subsequently used in the controller design. The structural response predicted by the plant predictor neural network is tested and assessed for wind excitations and its competence to predict the future response is evaluated and presented. The controller performance is assessed and evaluated. Additionally, the controller performance is compared to the conventional control methods such as linear Quadratic Gaussian (LQG) controller. Finally, the neuro-controller robustness and stability are examined and verified.

1

T. Kobori, Y. Inoue, K. Seto, H. Iemura and A. Nishitani (eds.), Proc. of Second World Conference on Structural Control, John Wiley & Sons, N.Y. (1998).

2

J.N. Yang, J.C. Wu, B. Samali, and A.K. Agrawal, "A Benchmark Problem for Response Control of Wind-Excited Tall Buildings", Proc., 2nd World Conf. On Structural Control, Vol.2, 1408-1416, (1999).

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £82 +P&P)