Efficient dual-synthetic-jets-based lift enhancement strategy generated by deep reinforcement learning

Distributed dual-synthetic jet actuators (DSJAs) are first integrated with a plain flap for improving the lift enhancement control ability. Two control methods (flow separation control and virtual gurney flap control) and their coordinated control are numerically investigated. A hundred-environment parallel deep reinforcement learning (DRL) framework integrating long short-term memory (LSTM) networks is first constructed and applied to optimize the control strategy with the objective of maximum lift enhancement effect. Notably, the parallel number is the highest in publications, indicating the best computing efficiency. Results show that the lift enhancement ratio of flow separation control (98.44%) is more than ten times that of virtual gurney flap control (9.05%) under the same energy consumption. Simultaneous operations of separation control and virtual gurney flap control could realize a lift enhancement ratio of 91.14%, which is instead less than that of only separation control. The optimal performance is not achieved by the simultaneous maximum-voltage activation of all DSJAs. The best control strategy, with a lift enhancement ratio of 115% (larger than the sum of the individual effect produced by separation control and virtual gurney flap control alone), is generated by DRL framework. Under control of this strategy, distributed DSJAs operate in a pulse mode and achieve a better lift enhancement effect with a higher effectiveness-cost ratio (2.46 times that of coordinated control) and lower energy consumption (less than half of coordinated control).