Interpretable Kepler-optimized ML for MEMS cantilever microswitch design

This study presents a novel hybrid machine learning framework that integrates the Kepler Optimization Algorithm (KOA) with Generalized Additive Models (GAM) to accurately predict the end displacement in MEMS cantilever-type microswitches. A parametric electro-mechanical coupling finite element model is developed to systematically generate displacement datasets across various parameter configurations, forming a robust foundation for model training and validation. Comparative experiments incorporating five machine learning models indicate that the put-forward KOA-GAM model significantly outdoes reference models in the aspect of prediction accuracy. The results demonstrate that KOA-GAM excels in capturing nonlinear displacement behaviour. By utilizing Shapley Additive Explanations (SHAP), the contributions of key design parameters to microswitch displacement are quantitatively analysed, uncovering their distinct influence mechanisms. The trained KOA-GAM framework offers an efficient alternative for rapid displacement prediction in cantilever-type microswitches, providing theoretical guidance for parameter control in microswitch design and facilitating structural optimization and performance improvement.