-
针对经典圆柱绕流问题, 采用深度强化学习方法, 提出了基于壁面压力反馈的圆柱绕流减阻闭环主动控制方法, 并比较分析了施加控制前后圆柱阻力系数、升力系数及流场的差异. 控制系统中, 以圆柱壁面上均匀分布的压力探针测得的信号作为反馈, 利用多层感知机建立压强信号与吹/吸射流及控制效果的映射关系, 即控制策略; 通过在圆柱上下表面狭缝施加连续可调的吹/吸射流来进行主动控制. 同时, 利用深度强化学习中的近端策略优化方法, 在大量的学习过程中对该控制策略进行不断调整和优化, 以实现稳定减阻效果. 在圆柱绕流流动环境搭建方面, 采用格子Boltzmann方法建立与深度强化学习模型之间的交互式框架, 模拟提取非定常流场条件下圆柱表面的压强信号, 并计算实时调整吹/吸射流强度时圆柱表面升力、阻力数据, 以评估所选控制策略的优劣. 研究表明: 雷诺数为100时, 主动控制策略能减少约4.2%的圆柱阻力, 同时减少约49%升力幅度; 同时施加主动控制后圆柱的减阻效果与圆柱回流区长度呈现强相关趋势. 此外, 不同雷诺数下智能体习得的策略减阻效果不同, 雷诺数为200和400时, 该主动控制策略能依次减小圆柱阻力17.3%和31.6%. 本研究可为后续开展基于壁面压力反馈的圆柱流动主动控制实验以及其他复杂环境下钝体流动智能控制提供参考.Focusing on the typical problem of flow around a circular cylinder, we propose an active flow control method of reducing drag of a circular cylinder, in which a deep reinforcement learning (DRL) method is used to establish the closed-loop control strategy with pressure sensors providing feedback signals. The detailed comparisons of the lift, drag, and flow fields with and without control are conducted. In the control system, pressure sensors evenly distributed on the cylinder surface are used to provide feedback signals for the controller. The multilayer perceptron is adopted to establish the mapping relationship between the sensors and the blowing/suction jets, i.e. the control strategy. A pair of continuously adjustable synthetic jets that exert transverse force mainly on the top and bottom edge of the cylinder is implemented. Based on the state-of-the-art proximal policy optimization algorithm, the control strategy is explored and optimized during a large number of learning episodes, thus achieving an effective, efficient, and robust drag reduction strategy. To build up the high-fidelity numerical environment, we adopt the lattice Boltzmann method as a core solver, which, together with the DRL agent, establishes an interactive framework. Furthermore, the surface pressure signals are extracted during the unsteady simulation to adjust the real-time blowing/suction jets intensity. The lift information and the drag information are recorded to evaluate the performance of the current control strategy. Results show that the active control strategy learnt by the DRL agent can reduce the drag by about 4.2% and the lift amplitude by about 49% at Reynolds number 100. A strong correlation between the drag reduction effect of the cylinder and the elongated recirculation bubble is noted. In addition, the drag reduction rate varies over a range of Reynolds numbers. The active control strategy is able to reduce the drag by 17.3% and 31.6% at Reynolds number 200 and 400, respectively. Owing to the fact that wall pressure signals are easy to measure in realistic scenarios, this study provides valuable reference for experimentally designing the active flow control of a circular cylinder based on wall pressure signals and intelligent control in more complicated flow environments.
-
Keywords:
- active flow control/
- flow past a circular cylinder/
- deep reinforcement learning/
- wall pressure feedback
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] -
Mesh T/δt $ \overline {{C_{\text{D}}}} $ $ \overline {\left| {{C_{\text{L}}}} \right|} $ Sr Ⅰ 768×144 1000 3.192 0.612 0.3026 Ⅱ 1536×288 2000 3.201 0.639 0.3019 Ⅲ 3072×576 4000 3.201 0.640 0.3012 
下载:
导出CSV
-
[1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29]
下载:
计量
- 文章访问数:5044
- PDF下载量:93
- 被引次数:0