Application of Artificial Neural Network (ANN) for predicting the behavior of micromachined diaphragm actuated electrostatically

In this study, a novel Artificial Neural Network (ANN) based on the feed-forward back-propagation (FFBP) algorithm has been used to predict the deflections of a rectangular diaphragm actuated electrostatically under different loadings and geometrical parameters. A limited range of simulation results obtained via CoventorWarereg will initially be used to train the neural network via back-propagation algorithm. The focus of this study would be to ease the process of parametric studies where the effects of varying the applied voltage, length, width, thickness, air gap and residual stress on the deflections of a polysilicon diaphragm will be investigated using ANN. Results obtained via ANN simulations are compared with results from CoventorWarereg simulations and existing analytical work for validation purpose. The proposed ANN model accurately predicts the deflections of the diaphragm with great reduction of simulations time and efforts, establishing the method superiority. The method can be extended to cases of cantilevers or fixed-fixed beams actuated through different excitation schemes and also for predicting other preferred parameters such as stroke volume and pull-in voltage.