The magnetorheological elastomer(MRE) is a smart material that has a special property of variable stiffness that is controllable by an externally applied magnetic field. Typical applications of MREs are tunable vibration absorbers and seismic isolation devices because the material stiffness can be easily controlled to tune the system at the optimal resonance frequency for vibration control. 

 

However, there exist some challenges to designing reliable and durable MRE based applications because the smart material has time-varying and nonlinear properties of stiffness variation. MRE samples produced under the same manufacturing process can have different stiffness values depending on external factors such as temperature, humidity and operation time. Since it is difficult to derive an accurate system model of an MRE, there is a limitation to obtain optimal and reliable performance using a model-based controller method.

 

Therefore, this study proposes a control method based on reinforcement learning(RL) without knowing the system modeling for a nonlinear MRE based vibration absorber system. A reinforcement learning control is a model-free method that learns the action of control in the direction to obtain the greatest reward under the state from the environment. For this study, the environment is the vibration absorber system, the state is the external vibration disturbance, the action is the magnetic field applied to the MRE, and the reward is the vibration reduction output of the absorber system. 

 

The proposed vibration control method defined the state and the reward from the signals of the accelerometers that are measuring the input disturbance and the output vibration of the system, respectively. The control action is transferred to the system by applying the calculated magnitude of the magnetic field. The proposed method is based on a simple architecture of ANN(artificial neural network) with 2 hidden layers and PPO(Proximal Policy Optimization) learning algorithm, which is suitable for real-time learning.

 

The designed RL method is applied on an MRE based tunable vibration absorber that is designed to suppress external disturbance of a linear compressor. The hardware has an embedded magnetic field generator which can generate the magnetic field up to 340 mT on the MREs. The magnetic generator is composed of permanent magnets placed on a BLDC flat motor and different magnetic field can be generated by controlling the relative position of the magnets from the MREs.

 

Experiments are conducted to evaluate the performance of the proposed method. First, sine-sweep modal tests were conducted, under different magnetic fields from 0mT to 340 mT, to derive the frequency response function(FRF) curves of the system. From analyzing the FRF results, we could manually estimate the optimal values of the magnetic field to maximize the vibration suppression in the target range from 45 to 65 Hz. Then, these estimated optimal values are compared with the resulting magnetic field action of the proposed RL method. Experiments showed that the proposed method generated action values similar to the optimal values from the manual estimation. Also, tests showed that the proposed control method could reduce the vibration as much as 42% compared to the state when the controller was turned off. Through the experiment, we could validate the feasibility of applying an RL control method without knowing the system modeling on an MRE based vibration absorber system.  


Feasibility Study of Reinforcement Learning Control for Magnetorheological Elastomer Based Tunable Vibration Absorber System

Division / Sponsors: 
Aerospace Division

Paper Proceedings: https://asmedigitalcollection.asme.org/SMASIS/SMASIS2020/volume/84027

SMASIS2020 - Virtual Conference

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.