A Hybrid Preprocessing Multi-Objective Surrogate Model for Thermal MEMS Actuators

In this study, an advanced surrogate model is proposed to simultaneously predict five key output variables, including deformation, stress, temperature, current density, and resonance frequency. This study used two models: Gaussian Process Regression (GPR) and an ensemble model based on Random Forest and XGBoost. By generating 10,000 design samples using the Latin Hypercube sampling method and performing simulations in COMSOL Multiphysics, as well as applying eight preprocessing methods, GPR achieved a mean absolute percentage error (MAPE) between 0.81% and 2.58%, whereas the ensemble model's MAPE ranged from 3.05% to 9.20%. The ensemble model offers substantially faster training, whereas GPR achieves higher prediction accuracy across all output variables. Additionally, a 5-fold cross-validation scheme was implemented to ensure reliable model evaluation. This surrogate model, achieving multi-objective prediction with strong scalability due to efficient preprocessing and sampling strategies, is an effective step in reducing computational costs and accelerating the design process of MEMS actuators.