Strategies for energy-efficient flow control leveraging deep reinforcement learning
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Physics. The day in general physics: instruments, methods and results across the field.
Kai: We're looking at the paper: "Strategies for energy-efficient flow control leveraging deep reinforcement learning".
Mira: Here's the gist of the paper. This study investigates active flow control in two-dimensional flows at a Reynolds number of 100 using Deep Reinforcement Learning (DRL) to develop flow control strategies that enhance energy efficiency and minimize energy consumption, thereby addressing the limitations of traditional methods The optimal jet placement for both square and circular cylinders is at the main flow separation point, achieving the best balance between energy efficiency and control effectiveness Flow Control Mechanism Active flow control (AFC) involves regulating and controlling fluid flow through external mechanisms such as injection, suction, or vibration to optimize flow characteristics and enhance system efficiency, performance, and stability The interaction of the synthetic jet with the flow generates new vortices that modify the local flow structure, significantly enhancing the cylinder’s aerodynamic performance For a circular cylinder at approximately 105° from the stagnation point, positioning requires only 1% of the inlet flow rate and achieves an 8% reduction in drag, with energy consumption one-third of that at other positions DRL Training Signal Construction The DRL agent interacts with the environment by taking actions based on its policy, which guides action selection from the current state and provides feedback as rewards The mathematical model of the reinforcement learning problem is defined by a finite Markov Decision Process (MDP), involving the state space S, action space A, transition probability P, reward R, and discount factor gamma The policy optimization algorithm employed is PPO, which uses a clipped objective to keep the probability ratio of the current and old policies within a safe range Performance Gains for Circular Cylinders The single-jet training results indicate that positions J1 and J5 show slower and lower reward function convergence, suggesting lower training effectiveness and energy efficiency In contrast, positions J2, J3, and J4 achieve rapid convergence and superior control strategy acquisition The most effective jet placements are at 75°, 90°, and 105°, where the wake of the cylinder is fully stabilized, and vortex shedding is completely eliminated Performance Gains for Square Cylinders For a square cylinder, placement at J3 yields the best performance, achieving 14.4% drag reduction with only 1.7% of the inlet flow rate—demonstrating both high control effectiveness and energy efficiency Configurations like J13, J23, which demonstrate rapid reward growth, are evidently more conducive to achieving efficient and precise flow manipulation The placement of the synthetic jets near the rear corners at position J3 provides more effective flow control, leading to the complete suppression of vortex shedding Energy Efficiency Insights The analysis shows that injectors positioned at 90° and 105° demonstrate superior efficiency, maintaining effective flow control with significantly lower energy consumption The configuration with jets located at J3 exhibits the highest peak in control effort, while EJ4 ≈ EJ2 The most energy-efficient option is J3, suggesting its optimal placement for long-term, low-energy operation DRL training results reveal that placing synthetic jets near 105° on a circular cylinder and at the rear corners of a square cylinder yields optimal energy efficiency and flow control performance Positioning jets at separation points enhances momentum exchange and induces beneficial vortex structures, promoting boundary layer reattachment without mass addition The effectiveness of DRL in flow control lies in its ability to autonomously learn optimal strategies through interaction with complex, nonlinear environments Strategic placement of synthetic jets allows for a favorable trade-off between control performance and energy efficiency Coordinated jet configurations yield synergistic effects, enhancing overall control efficacy The findings underscore the pivotal role of jet placement in achieving robust, energy-efficient flow control Appendix Summary The mesh independence study confirms that the mean drag coefficient, maximum drag coefficient, and maximum lift coefficient all exhibit small variations and converge towards consistent values with increasing mesh resolution The results are in close agreement with the reference benchmark study by Schafer et al., further validating the numerical setup The PPO algorithm demonstrates clear advantages over the Vanilla PG approaches in both training stability and flow control performance The PPO-based controller achieves the most rapid and pronounced drag reduction, with CD dropping below 2.98 The PPO-based strategy exhibits the fastest decay of CL oscillations, achieving nearzero mean fluctuations within a short time window The PPO algorithm facilitates faster policy acquisition, more effective drag suppression, and enhanced lift stabilization The study demonstrates that strategic placement of synthetic jets allows for a favorable trade-off between control performance and energy efficiency Coordinated jet configurations yield synergistic effects, enhancing overall control efficacy The findings underscore the pivotal role of jet placement in achieving robust, energy-efficient flow control The optimal position for synthetic jets around a circular cylinder is found to be at a 105° angle from the front stagnation point At this position, drag and lift are significantly reduced, vortex shedding is fully suppressed, and energy consumption is minimized The results further confirm that the 105° configuration offers the best balance between energy efficiency and flow stability The comparison between the baseline flow and the controlled flow, with jets positioned at the rear corners (J13 and J23), is illustrated in figure 20 The placement of the synthetic jets near the rear corners effectively suppresses vortex formation and stabilizes the wake flow The interaction between synthetic jets and the external flow induces significant changes in local flow patterns and streamlines, facilitating linear momentum transfer without mass addition The alternating phases of blowing and suction generate vortices that enhance boundary layer mixing, supplying the necessary energy to overcome adverse pressure gradients and enabling complete flow reattachment around the square cylinder The comparison between the baseline flow and the controlled flow, with jets positioned at 105° (J12 and J22), is illustrated in figure 19 Upon activation of the synthetic jets, as depicted in figure 19(b), the recirculation region on the leeward side of the cylinder expands significantly, and vortex shedding is completely suppressed The synthetic jets located at J3 maintain the highest sustained action values, indicating a greater continuous energy requirement for maintaining effective control The persistent energy efficiency of the J3 configuration across both two- and three-jet arrangements suggests that this location—likely corresponding to critical regions such as separation or reattachment points—offers an optimal balance between control authority and energy cost The comparison between the baseline flow and the controlled flow, with jets positioned at the rear corners (J13 and J23), is illustrated in figure 20 The placement of synthetic jets near the rear corners effectively suppresses vortex formation and stabilizes the wake flow The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The results crystallize the crucial role of optimal jet placement in flow control, where precise modulation of aerodynamic forces, such as lift, is essential for achieving effective performance The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities <ref:2410.
Paper summary: Mira: We just covered the setup of this paper "Strategies for energy-efficient flow control leveraging deep reinforcement learning." The core idea they are presenting is using Deep Reinforcement Learning to develop flow control strategies that enhance energy efficiency and minimize energy consumption by replacing traditional methods.
Kai: They claim that the optimal jet placement for both square and circular cylinders is at the main flow separation point because it achieves the best balance between energy efficiency and control effectiveness.
Lev: But what did they actually build in terms of hardware? Did they test this on a specific geometry, like a cylinder or a square shape?
Mira: They investigated two dimensional flows at a Reynolds number of one hundred using Deep Reinforcement Learning to develop flow control strategies that enhance energy efficiency and minimize energy consumption, thereby addressing the limitations of traditional methods <ref:2410.00424#pg1,to develop flow control strategies that enhance energy efficiency and minimize energy>.
Kai: The abstract points out that for the circular cylinder, positioning the jet at approximately one hundred five degrees from the stagnation point requires only one percent of the inlet flow rate and achieves an eight percent reduction in drag <ref:2410.00424#pg1,for the circular cylinder, positioning the jet at approximately 105>.
Mira: And they also found that at this same position for a circular cylinder, energy consumption is only one-third of what it would be at other positions.
Lev: So they’re using DRL to find these specific placement strategies, but what was the mathematical model underneath? What was the framework they used to define the problem?
Kai: The mathematical model is defined by a finite Markov Decision Process, which involves a state space S, an action space A, transition probability P, reward R and discount factor gamma.
Mira: And they use the PPO policy optimization algorithm to optimize this MDP using a clipped objective to keep the probability ratio of the current and old policies within a safe range.
Lev: From an engineering standpoint, that sounds computationally intensive. How does that DRL agent interact with the environment in practice? What’s its learning signal?
Kai: The DRL agent interacts with the environment by taking actions based on its policy, which guides action selection from the current state and provides feedback as rewards. This interaction is guided by a finite Markov Decision Process.
Mira: They found that for single synthetic jet training results, positions J1 and J5 show slower and lower reward function convergence, suggesting lower training effectiveness and energy efficiency in those spots.
Lev: That’s a clear limitation; if the reward isn't converging fast enough, the whole learning process stalls. How does that affect real hardware implementation?
Kai: In contrast, positions J2, J3 and J4 achieve rapid convergence and superior control strategy acquisition for single jet training results.
Mira: For square cylinders specifically, placement at position J3 yields the best performance, achieving fourteen point four percent drag reduction with only one point seven percent of the inlet flow rate.
Lev: Fourteen point four percent is a solid number for a drag reduction target, but what about the energy consumption comparison there? How does it stack up against other placements?
Kai: The paper shows that configurations like J13 and J23 demonstrate rapid reward growth, which are evidently more conducive to achieving efficient and precise flow manipulation for the square cylinder.
Mira: The overall innovation summarized is the sensitivity analysis of synthetic jet placement, highlighting DRL’s superior decision making in effective flow control, which is what makes this work interesting.
Lev: So they're emphasizing that it's not just about finding one good spot; it’s about how many jets you use and exactly where you put them to get the best result.
Kai: That’s right, and it shows that coordinated jet configurations can yield synergistic effects, which boosts the overall control efficacy beyond what a single jet placement could do.
Conclusion: Mira: So wrapping up this study on "Strategies for energy-efficient flow control leveraging deep reinforcement learning." The authors focused on developing these DRL strategies to make flow control more energy efficient and less wasteful than traditional methods.
Kai: The main finding they highlight is that the optimal jet placement for both square and circular cylinders is at the main flow separation point, which balances efficiency and effectiveness.
Lev: In simpler terms, what does this mean for someone who isn't deep into DRL? What’s the practical implication here for an engineer working on fluid dynamics?
Mira: It means that if you want to control a flow efficiently, you shouldn't just guess where to put your jets; the system needs to learn through interaction. The DRL learns the right configuration autonomously.
Kai: The authors are showing that strategic placement of synthetic jets enhances momentum exchange and induces beneficial vortex structures, promoting boundary layer reattachment without needing extra mass input.
Lev: So, what’s the biggest practical shift for someone who only cares about running this on real hardware? Are we talking about a simple software update or a complete redesign of the control mechanism?
Mira: It suggests that instead of pre-programming every possible situation, you let the DRL find those optimal trade-offs between control performance and energy cost.
Kai: The paper shows that coordinated jet configurations yield synergistic effects, which boosts overall control efficacy beyond what any single jet placement could do on its own.
Lev: So the final point is that for real hardware, you need to consider the overhead of running this learning agent versus the gains it provides in terms of drag reduction and stability.
Mira: That’s right, and it confirms that precise modulation of aerodynamic forces is essential for getting good performance in these flows.
Wang Jia, Hang Xu
State Key Laboratory of Ocean Engineering, School of Ocean and Civil Engineering, Shanghai Jiao Tong University
physics.flu-dyn
Submitted: 2024-10-01
Updated: 2025-07-21
Journal ref: Engineering Applications of Artificial Intelligence 156, 111125 (2025)
DOI: 10.1016/j.engappai.2025.111125
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 83/100
The gist: This study investigates active flow control in two-dimensional flows at a Reynolds number of 100 using Deep Reinforcement Learning (DRL) to develop flow control strategies that enhance energy…
Key concepts
- Active Flow Control (AFC)
- AFC involves using external mechanisms like injecting or sucking fluid to regulate flow characteristics. In this study, synthetic jets were used to generate new vortices that modify the local flow structure around cylinders, significantly improving aerodynamic performance and efficiency.
- Deep Reinforcement Learning (DRL)
- DRL is a machine learning technique where an agent learns optimal actions through interaction with an environment. The agent takes actions based on its current state, receives rewards or penalties, and adjusts its policy to maximize the reward over time, allowing it to discover complex flow control strategies autonomously.
- Markov Decision Process (MDP)
- An MDP is a mathematical framework used to model reinforcement learning problems. It defines the problem using states (the current situation), actions (what the agent can do), transition probabilities (how actions change the state), rewards, and a discount factor, which guides long-term decision-making.
- Policy Optimization Algorithm (PPO)
- PPO is a specific algorithm used in DRL for policy optimization. It uses a clipped objective to ensure that the agent's policy updates are stable and safe. This method helps the DRL agent learn faster and more effectively than traditional reinforcement learning approaches.
Terminology
Summary
This study investigates active flow control in two-dimensional flows at a Reynolds number of 100 using Deep Reinforcement Learning (DRL) to develop flow control strategies that enhance energy efficiency and minimize energy consumption, thereby addressing the limitations of traditional methods The optimal jet placement for both square and circular cylinders is at the main flow separation point, achieving the best balance between energy efficiency and control effectiveness
Flow Control Mechanism
Active flow control (AFC) involves regulating and controlling fluid flow through external mechanisms such as injection, suction, or vibration to optimize flow characteristics and enhance system efficiency, performance, and stability The interaction of the synthetic jet with the flow generates new vortices that modify the local flow structure, significantly enhancing the cylinder’s aerodynamic performance For a circular cylinder at approximately 105° from the stagnation point, positioning requires only 1% of the inlet flow rate and achieves an 8% reduction in drag, with energy consumption one-third of that at other positions
DRL Training Signal Construction
The DRL agent interacts with the environment by taking actions based on its policy, which guides action selection from the current state and provides feedback as rewards The mathematical model of the reinforcement learning problem is defined by a finite Markov Decision Process (MDP), involving the state space S, action space A, transition probability P, reward R, and discount factor gamma The policy optimization algorithm employed is PPO, which uses a clipped objective to keep the probability ratio of the current and old policies within a safe range
Performance Gains for Circular Cylinders
The single-jet training results indicate that positions J1 and J5 show slower and lower reward function convergence, suggesting lower training effectiveness and energy efficiency In contrast, positions J2, J3, and J4 achieve rapid convergence and superior control strategy acquisition The most effective jet placements are at 75°, 90°, and 105°, where the wake of the cylinder is fully stabilized, and vortex shedding is completely eliminated
Performance Gains for Square Cylinders
For a square cylinder, placement at J3 yields the best performance, achieving 14.4% drag reduction with only 1. p. 07% of the inlet flow rate—demonstrating both high control effectiveness and energy efficiency Configurations like J13, J23, which demonstrate rapid reward growth, are evidently more conducive to achieving efficient and precise flow manipulation The placement of the synthetic jets near the rear corners at position J3 provides more effective flow control, leading to the complete suppression of vortex shedding
Energy Efficiency Insights
The analysis shows that injectors positioned at 90° and 105° demonstrate superior efficiency, maintaining effective flow control with significantly lower energy consumption The configuration with jets located at J3 exhibits the highest peak in control effort, while EJ4 ≈ EJ2 The most energy-efficient option is J3, suggesting its optimal placement for long-term, low-energy operation
Conclusion on Optimal Placement
DRL training results reveal that placing synthetic jets near 105° on a circular cylinder and at the rear corners of a square cylinder yields optimal energy efficiency and flow control performance Positioning jets at separation points enhances momentum exchange and induces beneficial vortex structures, promoting boundary layer reattachment without mass addition The effectiveness of DRL in flow control lies in its ability to autonomously learn optimal strategies through interaction with complex, nonlinear environments Strategic placement of synthetic jets allows for a favorable trade-off between control performance and energy efficiency Coordinated jet configurations yield synergistic effects, enhancing overall control efficacy The findings underscore the pivotal role of jet placement in achieving robust, energy-efficient flow control
Appendix Summary
The mesh independence study confirms that the mean drag coefficient, maximum drag coefficient, and maximum lift coefficient all exhibit small variations and converge towards consistent values with increasing mesh resolution The results are in close agreement with the reference benchmark study by Schafer et al., further validating the numerical setup The PPO algorithm demonstrates clear advantages over the Vanilla PG approaches in both training stability and flow control performance The PPO-based controller achieves the most rapid and pronounced drag reduction, with CD dropping below 2.98 The PPO-based strategy exhibits the fastest decay of CL oscillations, achieving nearzero mean fluctuations within a short time window The PPO algorithm facilitates faster policy acquisition, more effective drag suppression, and enhanced lift stabilization The study demonstrates that strategic placement of synthetic jets allows for a favorable trade-off between control performance and energy efficiency Coordinated jet configurations yield synergistic effects, enhancing overall control efficacy The findings underscore the pivotal role of jet placement in achieving robust, energy-efficient flow control The optimal position for synthetic jets around a circular cylinder is found to be at a 105° angle from the front stagnation point At this position, drag and lift are significantly reduced, vortex shedding is fully suppressed, and energy consumption is minimized The results further confirm that the 105° configuration offers the best balance between energy efficiency and flow stability The comparison between the baseline flow and the controlled flow, with jets positioned at the rear corners (J13 and J23), is illustrated in figure 20 The placement of the synthetic jets near the rear corners effectively suppresses vortex formation and stabilizes the wake flow The interaction between synthetic jets and the external flow induces significant changes in local flow patterns and streamlines, facilitating linear momentum transfer without mass addition The alternating phases of blowing and suction generate vortices that enhance boundary layer mixing, supplying the necessary energy to overcome adverse pressure gradients and enabling complete flow reattachment around the square cylinder The comparison between the baseline flow and the controlled flow, with jets positioned at 105° (J12 and J22), is illustrated in figure 19 Upon activation of the synthetic jets, as depicted in figure 19(b), the recirculation region on the leeward side of the cylinder expands significantly, and vortex shedding is completely suppressed The synthetic jets located at J3 maintain the highest sustained action values, indicating a greater continuous energy requirement for maintaining effective control The persistent energy efficiency of the J3 configuration across both two- and three-jet arrangements suggests that this location—likely corresponding to critical regions such as separation or reattachment points—offers an optimal balance between control authority and energy cost The comparison between the baseline flow and the controlled flow, with jets positioned at the rear corners (J13 and J23), is illustrated in figure 20 The placement of synthetic jets near the rear corners effectively suppresses vortex formation and stabilizes the wake flow The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The results crystallize the crucial role of optimal jet placement in flow control, where precise modulation of aerodynamic forces, such as lift, is essential for achieving effective performance The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effective in fully eliminating wake instabilities The comparison between the baseline flow and the controlled flow, with jets positioned at 15° (J1 on front face), their interaction with the incoming free stream improves near-wake stability; however, alternating vortex shedding persists, suggesting that upstream jet placement is less effectiv… p. 5
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
System can autonomously learn optimal jet placement by minimizing energy consumption while maximizing drag reduction, as demonstrated by finding that
the optimal jet placement for both square and circular cylinders is at the main flow separation point.
-
The system can achieve complete suppression of vortex shedding and stabilize flow without oscillations, as the DRL-based strategy
completely eliminates vortex shedding and fully stabilizes the flow,
contrasting with previous studies that reported persistent instabilities. -
The system can intelligently select multiple jet configurations based on predicted energy efficiency, specifically choosing configurations like
the configuration with jets located at J2 demonstrates the highest energy efficiency
when controlling a square cylinder. -
The system can dynamically adapt its control strategy in real-time without needing predefined parameters, as it
removes the requirement for predefined control parameters, allowing the agent to autonomously infer effective strategies from flow feedback.
-
The system can provide precise guidance on optimal jet placement by correlating physical phenomena with DRL training results, such as identifying that
placing synthetic jets near 105° on a circular cylinder and at the rear corners of a square cylinder yields optimal energy efficiency and flow control performance.
Sources
- Playing Atari with Deep Reinforcement Learning
- Deep Reinforcement Learning: An Overview
- Proximal Policy Optimization Algorithms
- Applying deep reinforcement learning to active flow control in turbulent conditions
- Reinforcement-learning-based control of confined cylinder wakes with stability analyses
- Active flow control for drag reduction through multi-agent reinforcement learning on a turbulent cylinder at $Re_D=3900$
Related papers
- Emergent Transfer of a Physics Foundation Model from Simulation to Laboratory Turbulence
- Early onset of secondary shear instability in Kelvin-Helmholtz braids at high Reynolds number
- Rotational covariance, continuous spectrum and non-modal growth of three-dimensional planar weak viscous shock layers
- Deep Learning-based Algebraic Reynolds Stress Closures for RANS Simulations of Turbulent Flows
- Neptuna: A Comprehensive Machine Learning Framework for Benchmarking Complex Multiphase Flows
- Size and shape of condensing steam jets and plumes