Analyzing the geometric dependence of thermoelastic Q-factor in micro hemispherical resonators via a data-augmented...

Area of Science:

• Mechanical Engineering
• Materials Science
• Artificial Intelligence

Background:

• The thermoelastic damping Q-factor (Q_TED) is crucial for Micro Hemispherical Resonators (MHRs).
• Finite-element analysis (FEA) for Q_TED prediction is computationally expensive and time-consuming.

Purpose of the Study:

• To develop a rapid and accurate prediction framework for Q_TED in MHRs.
• To optimize MHR design for enhanced Q_TED by analyzing geometric parameter influences.

Main Methods:

• A hybrid Convolutional Neural Network (CNN)-Transformer model was developed.
• Data augmentation using polynomial fitting of FEA results enhanced model accuracy.
• Ablation studies identified an optimal architecture with multi-head self-attention mechanisms.

Main Results:

• The proposed AI model significantly outperforms existing prediction methods in accuracy and robustness.
• Optimal MHR design for maximum Q_TED involves low height (H), low anchor radius (r), and high thickness (T).
• A low anchor radius (r) is recommended to mitigate fabrication sensitivity.

Conclusions:

• The developed framework offers a computationally efficient and reliable alternative to FEA for MHR design.
• The study provides insights into geometric parameter optimization for maximizing Q_TED in MHRs.
• This AI-driven approach facilitates the optimization and robust design of high-performance MHRs.