Estimating Effects of Residual Physics with Machine Learning for Earbud Performance Prediction

Consumer electronics devices like earbuds include small components like the voice coil speaker. The damping coefficient of the voice coil speaker system and the acoustic resistance of the mesh at the aperture used in Finite Element Analysis (FEA) based simulation needs to be tuned to match predictions with measurements. In this paper, a machine learning method is proposed to combine experimental data with simulation data, that accounts for uncertainties in model and parameters to obtain a hybrid analytics model that can predict the performance of the speaker for unmeasured parameter values.