Akademik Veri Yönetim Sistemi
Journal of Computational Electronics, cilt.25, sa.1, 2026 (SCI-Expanded, Scopus)
Displacement and naturel frequency are the most important design parameters for diaphragms based microelectromechanical system (MEMS) pressure sensors. For nonconventional diaphragm design of MEMS devices, finite element method (FEM)-based analysis to obtain these two parameters requires quite long time and cost as compared to conventional diaphragm design including circular, square, and rectangular shape. Thus, one major disadvantage of FEM is the excessive time required for simulation. Machine learning (ML) algorithms might be an alternative approach to FEM analysis. ML algorithms, which is an easier, functional, and time and cost saving, might provide rapid prediction of essential information comprising displacement and naturel frequency of MEMS diaphragm design with accurate and reliable results. In this study, ML algorithms including XGBoost regressor, LightGBM regressor, CatBoost regressor, and TabNet regressor were used to estimate displacement (µm) and frequency (Hz) of perforated low temperature co-fired ceramic (LTCC) diaphragms using 200 FEM-based numerical results. Predicted results were compared by considering R2, MAE, RMSE, and MAPE metric. According to these results, best performance was obtained by CatBoost regressor with the values of R2 = 0.927 and R2 = 0.995 for the displacement and frequency prediction, respectively. It was realized that CatBoost strikes an exceptional balance between computational efficiency and predictive performance, while LightGBM emerges as a strong alternative for scenarios prioritizing speed and memory efficiency. As a result, it was concluded that ML algorithms might be a useful, cost, and time effective tools for rapid analysis of displacement and naturel frequency of perforated diaphragms without requiring FEM analysis.