Machine Learning Based Nonlinear Resonator System Identification

We use genetic algorithm (GA) based machine learning to extract the system parameters of a nonlinear MEMS/NEMS response. The fitting is carried out using the standard analytical nonlinear response of a 1D system as a fitting function without using data from the linear response as a starting point. We demonstrate that the extracted values for parameters such as driving force, effective mass and capacitive sensing gain closely match the values obtained using standard analytical models. More importantly, we extract values for Quality factor and nonlinear coefficients which are not readily calculated and recover the non-linear as well as linear frequency response with maximum normalized mean square error (NMSE) of 0.02. Through this work, we unearth the amplitude dependence of quality factor in our resonators as well as the temperature dependence of nonlinear parameters.