Aerodynamic force and yaw moment control of a generic light truck using machine learning
Final Report Abstract
This research project investigated the active drag reduction of bluff bodies using two experimental setups and a multitude of optimization algorithms for open- and closed-loop control. Bluff bodies are omnipresent in the engineering field, and the desire to reduce the acting drag is high. The project research activities were enabled by an innovative design and construction of a truck model attached to a yawing mechanism and equipped with 7 pneumatic actuators and 18 fast pressure sensors. Each actuator was capable of unsteady blowing at 3 blowing intensity levels. This rich set of actuators and sensors allowed an unprecedented survey of the broad range of actuation possibilities. This, in turn, was enabled by development of open- and closed-loop algorithms which identified the optimal actuation strategy in this high dof problem. The flow analysis and modeling of the actuated D-shaped bluff body revealed the flow physics responsible for drag response. Distinct regions of synchronization and desynchronization were identified and accurately modeled. The model is a modified Stuart−Landau equation with a general forcing term. The modeling as well as the analyses was enabled by a spectral analysis across a range of actuation frequencies and blowing intensities; Tools that can be applied to a broad range of actively controlled flow problems. Experimental and numerical investigations on the truck model revealed several new observations; Coanda actuators along the front edge are only beneficial when the flow is separated, i.e., on models with sharp corners along the leading edge. Moreover, actuation along the vertical front portion of the model is more energy-efficient than over the upper part (A-pillar) or entire front edge. Furthermore, steady Coanda actuation on the model base is favored for actuators with relatively large Coanda radius. This finding was verified using both open- and closed-loop control. Initial results from the latest measurement campaign using smaller Coanda radiuses have shifted this inclination toward unsteady blowing. A clearer picture on that point is expected to immerge as we conclude the data analysis from the last measurement campaign. The achieved results were enabled by algorithmic development in both open- and closed loop control. Two optimization algorithms were deployed on the experiment: the Explorative Gradient Method (EGM), and the Bayesian optimization algorithm. These algorithms were capable of significantly reducing the optimization time in this very large degrees-of-freedom problem. Machine learning control supplemented with gradient enrichment was also deployed on the experiment to minimize the cost function under steady and unsteady conditions. The ability to identify an optimized control law under in an automated and model-free manner under unsteady conditions is a unique contribution from this project. The project findings open new possibilities to further explore the subject of active drag control on bluff bodies both on the fundamental as well as the applied level. The developed models and algorithms are scalable and transferrable to other aerodynamic problems and applications.
Publications
-
"Open-and closed loop control on a D-shaped bluff body equipped with Coanda actuation." AIAA Paper 2019-3601, 2019
Oswald, P., Semaan, R., and Noack, B. R.
-
“Modeling synchronization in forced turbulent oscillator flows.” Nature Communications Physics 3(1), 2020: 1-9
Herrmann, B., Oswald, P., Semaan, R., & Brunton, S. L.
-
“Drag reduction of a D-shaped bluffbody using linear parameter varying control.” Physics of Fluids, 33(7), 2021, 077108
Shaqarin, T., Oswald, P., Noack, B. R., & Semaan, R.
-
“Stabilization of the fluidic pinball with gradient-enriched machine learning control.” Journal of Fluid Mechanics, 917, 2021
Maceda, G. Y. C., Li, Y., Lusseyran, F., Morzyński, M., & Noack, B. R.
-
“Explorative gradient method for active drag reduction of the fluidic pinball and slanted Ahmed body.” Journal of Fluid Mechanics, 932, 2022
Li, Y., Cui, W., Jia, Q., Li, Q., Yang, Z., Morzyński, M., & Noack, B. R.
