Optimizing the performance of electrostatic comb-drive actuators using neural networks

This paper investigates the application of two different meshing techniques used by the finite element method to analyze Micro-Electro-Mechanical Systems (MEMS) or structures. In this work, particular interest is focused on the application of these two techniques to comb-drive actuators, where a parametric study is carried out to optimize the design parameters of a lateral Electrostatic comb-drive actuator. In this case, the thickness, gab size, applied driving voltage and the number of Comb fingers are varied. The two meshing techniques applied are commonly known as, the Exposed Face Meshing method (EFM), and the Volume Refining Meshing (VRM) method. On one hand, the EFM algorithm permits the independent refinement of the electrostatic and the mechanical meshes while keeping full compatibility between the two meshed domains. On the other hand, the VRM method requires both, the electrostatic and the mechanical meshes, to be refined together at the same time; which result in a large problem size for complex structures with high aspect ratio. While the VRM method requires meshing of air gaps between the conductors, the EFM method uses the boundary element technique to map and calculate the electrostatic forces on the surfaces of the structure conductors, which eliminates the need to mesh the air gaps between the conductors, which is proved to be more efficient. In the current work, a calibration tool that can be used to enhance the efficiency of FE results produced by the VRM technique is developed. In this process, the results obtained from both meshing techniques of the para-metrically studied comb-drive are used to train a Feedforward Backpropagation Neural Network. This network is constructed using the FE results obtained from the (EFM) as the training target or output. The result obtained from the (VRM) technique is then fed as input to the network, which can then be calibrated using the EFM results.