Intelligent control systems for wind-induced vibration mitigation in tall buildings

Abstract

Wind-induced vibration (WIV) is a major contributor to the high investment cost of tall buildings. WIV can render a building unserviceable and strongly affect occupant comfort. It can also damage internal building components and may even cause catastrophic failure in severe cases. According to most design codes, tall buildings must be designed to endure severe wind events with return periods as long as 50 years, which requires additional construction material and design effort. Tuned mass/liquid dampers (TMDs/TLDs) have been designed and employed to reduce WIV effects. They rely on using energy dissipation mechanisms, such as oil flow through constrictions, to absorb WIV energy from the host structure. TMDs and TLDs are narrowband devices which are not adaptable to different wind characteristics. They also tend to consume space in higher floors where real estate is usually more expensive. To inhibit even the initial formation of WIV, aerodynamic modifications (AMs) have been used both independently and in conjunction with TMDs and TLDs. By changing the external shape of the building, the wind system attacking the building can be altered so the building becomes less susceptible to WIV. Major AMs change the overall shape of the building by introducing tapers, setbacks and helix features. Alternatively, minor AMs introduce corner or edge changes such as rounding, fillets or cut-offs. The literature review presented herein concludes that AMs have the potential to reduce WIV accelerations by 30-60%. However, AMs that are effective at reducing WIV accelerations at certain wind conditions may inadvertently increase those accelerations depending on the wind characteristics and the surrounding environment. This work focuses on designing and optimally controlling active aerodynamic modification systems that can change their orientation to guarantee consistent WIV reduction across different wind conditions and environmental changes. A central aim to the designs and techniques presented in this work is to generate versatile controllers that can work across different building and AM designs, wind characteristics and surrounding environments while minimizing user input and deployment effort. To achieve these goals various machine learning models and techniques are used because they generally support a black box design approach. An AM design using four plates installed on the corners of the building is given special attention in this work. The plates can rotate and hold position around their own axes of symmetry. This design is emphasized for its simplicity, ease of implementation and the minimal number of control parameters. However, the techniques developed herein can also be used for different AM designs. To simulate the WIV response at different wind conditions and plate orientations, this work uses (1) a fluid-solid-interaction simulation model developed specifically to enable plate rotation and building motion in the simulation domain, (2) wind tunnel experiments of a 1:400 scaled robotic aeroelastic building model and (3) large-scale wind tunnel experiments of a 1:73 scaled robotic aeroelastic building model. These models are controlled using (1) a wind condition averaging control approach using a developed iterative optimum training technique proven herein to use a substantially reduced number of experiments to obtain optimal training and (2) The combination of a distributed sensor network and recurrent neural networks to fully define the instantaneous system state using time series inputs. The final version of the control scheme developed in this work uses a genetic reinforcement learning method that enhances the exploration capacity of reinforcement learning and makes it more likely to find global optimum solutions. The results indicate that utilizing the developed designs and control techniques can reduce WIV response by 40-90% depending on the wind condition. The developed controller was able to minimize WIV in randomly varying wind conditions even for conditions it did not specifically encounter during training.