Vibration energy harvesting performance of a tuned inertial mass electromagnetic transducer controlled by deep reinforcement learning

Abstract Vibration energy harvesting technology utilizing electromagnetic transducers (ETs) from vibrating structures has garnered significant attention as a ubiquitous renewable energy source. Traditionally, the damping value induced by the motor of these devices for power generation is determined based on the impedance matching at a specific frequency. However, in environments with disturbances of varying frequencies, a control algorithm for variable damping is required to enhance power generation efficiency. Meanwhile, a tuned inertial mass electromagnetic transducer (TIMET) has been proposed to improve power generation efficiency by using a mechanism known as an inerter. This device achieves higher efficiency within a specific frequency range by resonating the inerter with a tuning spring. Therefore, it is necessary to expand the frequency range over which these devices operate efficiently. In this study, deep reinforcement learning (DRL) is employed to control motor damping in real time in response to disturbances of varying frequencies, thereby improving the power generation efficiency. More specifically, Double Deep Q-Network (DDQN), an improved version of Deep Q-Network (DQN), which is considered one of the most successful DRL methods, is applied to both ET and TIMET. Moreover, this study investigates the effectiveness of not only variable damping but also the utilization of regenerative energy. Initially, the DDQN agent is used to train policies for controlling the motor damping of the devices attached to a single-degree-of-freedom (SDOF) model to maximize power generation when a sinusoidal sweep wave is input. Subsequently, numerical simulations are conducted to assess the performance of the trained policies under other unknown disturbances with varying frequencies. The results obtained show that the proposed strategy can increase the power generation efficiency for disturbances of various frequencies.