Eddy current damper model identification using hybrid convolutional and recurrent neural network

The eddy current damper is an energy dissipating device, one of the main advantages of which over conventional fluid dampers include the ability to produce resistive forces with no contact between the components wherein damping forces are generated, resulting in a less degradable device with less maintenance requirements. However, one of the challenges in its development process is identifying the human-interpretable model, the mathematical law, which describes its behavior. Therefore, in this paper several existing approaches to address this issue are discussed along with their advantages and disadvantages, and the new method involving the usage of a hybrid convolutional (CNN) and recurrent (RNN) neural network is presented. As a machine learning model deals with data, the approach of the test data preparation and mathematical equation representation, as well as the machine learning model training process, is discussed. Finally, the performance of the trained model is evaluated using the eddy current damper experiment data, showing its capability to identify the mathematical model even in presence of noise, and the conclusion on the effectiveness of the proposed approach is made.