Nonlinear Vibration Control Of Large Space Membrane Structure Based On Deep Reinforcement Learning

Large-scale membrane structure snout is more and more widely used in the field of aerospace,which has the advantages of light mass,low cost,large area,small folding volume and high reliability.However,because of the vibration and shock in the space environment,the inertial force brought about by the space vehicle orbit ingress,and other factors,the large-size membrane is difficult to reduce vibration,affecting the membrane itself and the normal work of spacecraft,so how to alleviate the vibration of the membrane has attracted wide attention and research.Starting with the membrane boundary vibration control,this study established a finite element model of large-size membrane free vibration by ABAQUS software.With the aim of dissipating the vibration energy of the membrane,the vibration control of the membrane is carried out by observing the actuation mode of the amplitude of the boundary point.The control simulation of the velocity negative feedback mode was carried out by using the sub-program interface VUAMP of the membrane model.The validity of the dynamic method was verified,and the layout of the dynamic point was ulteriorly optimized.The simulation results show that this control method can effectively suppress the membrane vibration.Large-size membrane vibration has a strong nonlinearity,the establishment of dynamic model can be hardly applied to practical control because of its inaccuracy and the difficulty to solve the equations,but intensive learning methods can cover these shortages.This research project tries to transforms the control of membrane vibration into reinforcement learning issues,using the python language combined with the Tensor Flow framework to build a deep-reinforcement learning agent.This agent could gradually master the strategy of controlling thin membrane vibration through training,learning and exploration.In order to verify the effectiveness of the reinforcement learning control method,it is necessary to establish a ground simulation test system,which is designed with system framework,program and communication,so that the system can achieve highly automated experiments,and the reinforcement learning agent can become a membrane vibration suppression expert after this process.Finally,this study uses the hypothetical modal method to establish a finite element model of the membrane,then carry out a simulation on the multi-step vibration type of the model based on the reinforcement learning control method.It turned out to be that this method can achieve the expected result of efficient control after the completion of the agent training.Comparing the enhanced learning method with the velocity negative feedback method under the same conditions to control the vibration,the simulation results show its huge advantages including high efficiency at all periods of vibration control,thus showing the advantages of the reinforcement learning method.