Modeling Voice Coil Actuators with Recurrent Neural Network

In order to accurately model the behaviors of a voice coil actuator (VCA), three-dimensional (3-D) simulation is preferred over a lumped model. However, the 3-D simulation for a VCA can be very computationally expensive, which is due to the spatial discretization, the multiphysics nature, and the nonlinearities of the VCA system. In this work, we propose incorporating the recurrent neural network (RNN), to be specific, the long short-term memory (LSTM) into the multiphysics simulation. We solve the multiphysics problem by using the finite element method (FEM) at full 3-D accuracy in only a portion of the required time steps. We train a LSTM with the obtained FEM solution. Once the training completes, we replace the multiphysics simulation with the LSTM, which can make predictions and generate results on the remaining portion of the required time steps. With the proposed approach, we avoid solving nonlinear systems associated with the multiphysics problem repeatedly at all time steps and achieve a significant reduction of computation time.