Topology and shape optimization of piezoelectric ceramics using machine learning surrogate models trained on high-resolution XFEM simulations

The article proposes a topology and shape optimization hybrid of piezoelectric ceramics, which consists of high-resolution Extended Finite Element Method (XFEM) simulations alongside machine learning surrogate modeling. This proposed methodology has been identified to overcome the drawbacks associated with conventional finite element optimization, which is numerically intensive, have a hard time with geometries that are changing, and cannot easily capture the adverse effects of complex electromechanical couplings. XFEM uses the true electromechanical response, structural deformations, crack sensibility and field interactions under a wide range of varying geometry and loads in order to produce a high-fidelity collection of simulation data. According to this dataset, the sophisticated machine learning surrogates are trained to achieve important performance metrics such as the electromechanical coupling coefficient, output voltage, compliance, and equivalent stress with almost XFEM accuracy. Such surrogates allow exploration of design space in real time, and multi-objective optimization with high efficiency at a greatly lowered cost. Hybridized piezoelectric designs exhibit a high rate of electromechanical efficiency, lower compliance and material consumption, homogeneous stress distributions, and increased fatigue life. Robustness testing is a test of strong behavior in unproven circumstances, and convergence testing of good performance of efficient optimization.