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

Area of Science:

• Mechanical Engineering
• Computational Science
• Materials Science

Background:

• Micro-Electro-Mechanical Systems (MEMS) actuators require complex simulations for design.
• Predicting multiple performance metrics simultaneously is computationally intensive.
• Efficient surrogate modeling can accelerate the MEMS design cycle.

Purpose of the Study:

• To develop and compare advanced surrogate models for multi-objective prediction of MEMS actuator performance.
• To evaluate the accuracy and efficiency of Gaussian Process Regression (GPR) and an ensemble model (Random Forest and XGBoost).
• To assess the impact of sampling and preprocessing strategies on model scalability and computational cost.

Main Methods:

• Generation of 10,000 design samples using Latin Hypercube sampling.
• COMSOL Multiphysics simulations for data generation.
• Implementation and comparison of Gaussian Process Regression (GPR) and an ensemble model (Random Forest, XGBoost).
• Application of eight preprocessing techniques and 5-fold cross-validation.

Main Results:

• GPR achieved a Mean Absolute Percentage Error (MAPE) between 0.81% and 2.58% for five key output variables.
• The ensemble model exhibited a MAPE ranging from 3.05% to 9.20%.
• GPR demonstrated superior prediction accuracy, while the ensemble model provided faster training times.

Conclusions:

• The proposed surrogate model effectively predicts multiple MEMS actuator performance metrics.
• GPR offers higher accuracy, making it suitable for precise design predictions.
• The study demonstrates a significant reduction in computational costs and acceleration of the MEMS actuator design process.