<p>The growing demand for energy has led to significant attention being given to the energy harvesting process from vibrations using piezoelectric materials. Given the limited energy available for conversion, robust designs that minimize sensitivity to parameter uncertainties or external variations are essential. To ensure project quality, multi-objective optimizations are necessary to maximize the mean and minimize the standard deviation of the response, but the computational cost increases with the number of uncertain parameters, requiring more efficient approaches. In this way, with metamodels, which are computational tools, it is possible to provide a faster and less costly evaluation of such computationally expensive models. This study proposes the use of a Kriging metamodel to design robust cantilever beam energy harvesting devices, combined with Monte Carlo Simulation to estimate the mean and standard deviation of the Frequency Response Function of power output. Multi-objective optimization and sensitivity analysis are applied. Results indicate that using more design variables leads to a metamodel with higher computational cost due to the larger number of experimental samples required. Nevertheless, this cost remains low compared to direct model optimization, with a satisfactory time reduction in the optimization process.</p>