10:20
Classical and Machine methods in flow control I
Prediction of airfoil noise from velocity fields by artificial intelligence
Yannian Yang
Abstract: Airfoil tonal noise is known to be a result of the vortex shedding near the trailing edge, which is generated by the acoustic feedback loop. The unsteady flow field near the trailing edge is non-linearly related to the far field sound pressure, which can be learnt from the artificial intelligence(AI) network. In this study, a novel AI architecture is proposed to predict an accurate far field sound pressure from high-resolution particle image velocimetry data. As the ground truth, the sound pressure is measured by a far field microphone. A generative adversarial network (GAN) was employed for the first time in this problem. It is a
spectral-norm-based residual conditional GAN translator that provides a stable adversarial training and high-quality sound pressure generation. This network has the potential to predict sound pressure from velocity measurements, which always requires an anechoic chamber.
|
Active control of pitching airfoil flow based on deep reinforcement learning
Shixiong Zhang, Honglei Bai
Abstract: Dynamic stall can be frequently encountered in the aerospace engineering applications e.g., rotorcraft, small manned aerial vehicles (MAVs), and pitching airfoils. We present an experimental work on the active control of the flow over a pitching NACA0012 airfoil, based on deep reinforcement learning (DRL), aiming to improve the control effect and seek the optimal control parameters. The experiments are conducted in a low-speed wind tunnel at Reynolds number Re = U_∞c/ν = 1.6 × 105, where U_∞ is incoming flow velocity, c is the chord length of the airfoil and ν is the kinematic viscosity of air. The reduced frequency k = πfc/U_∞ is 0.075, where f is the oscillation frequency of the pitching airfoil. The function of the pitching motion is defined by α(t) = 10+15sin(2πft), where α is the angle of attack and t is time. The pulsation jet with 0.7%c width covering 73% of spanwise length of the airfoil is located at 5%c from the leading edge with an angle of 45° relative to the chord line. The control parameters include the driving pressure P_j^* (=P_j/P_a, where P_a is the atmospheric pressure), jet frequency F_j^* (= F_j X_e/U_∞, where X_e is the distance between the exit of the jet and the trailing edge of the airfoil) and duty cycle of the jet DC. The results show that the time-mean lift coefficients C ̅_L of the airfoil increase significantly. The maximum C ̅_L is up to 0.78 at step = 577, increasing by 59% of C ̅_Lo. The corresponding optimal control parameters are P_j^* = 2.31, F_j^* = 0.61 and DC = 58.5%. However, the convergence of F_j^* is not as well as the other parameters, which may be determined by the reward functions and the parameters of the DRL. It is indicated that the maximum lift coefficient is increased and the dynamic stall angle is delayed to 18° by the DRL-based control. The lift coefficient at α > 18° in the upstroke and in the down stroke increases significantly.
|
Model-Free Displacement Controller for a Tap-Water-Driven Artificial Muscle with Fuzzy Logic-Based Time-Varying Parameter Design
Ayaka Kosugi, Satoshi Tsuruhara, Kazuhisa Ito
Abstract: McKibben-type artificial muscle has several advantages for practical application; however, it is difficult to control with high precision due to its hysteresis characteristics. For this problem, model-free control (MFC) is a data-driven controller that can handle such a nonlinear system. On the other hand, it shows significant performance deterioration with rapid changes in reference value. In this study, to solve this problem, a time-varying parameter update law, which is a combination of MFC and fuzzy logic, is proposed in this study. The updated parameter plays an important role in MFC, which is a denominator of the control input equation. While the conventional adaptive control method has been widely known to have lower control performance in the transient response, the proposed method allows for designing update laws heuristically, and it is confirmed that it suppresses the overshoot, and maintains the performance in the steady-state response via experimental validation using the tap-water-driven artificial muscle.
|
|