Model-based Fault Detection and Identification (FDI) for prognostics rely on the comparison between the response of the monitored system and that of a digital twin. Discrepancies among the behavior of the two systems are analyzed to filter out the effect of uncertainties of the model and identify failure precursors. A possible solution to identify faults is to leverage a model able to simulate faults: an optimization algorithm varies the faults magnitude parameters within the model to achieve the matching between the responses of the model and the actual system. When the algorithm converges, we can assume that the fault parameters that produce the best match between the system and its digital twin approximate the actual faults affecting the equipment. The choice of an optimization algorithm appropriate for the task is highly problem dependent. Algorithms for FDI are required to deal with multimodal objective functions characterized by poor regularity and a relatively high computational cost. Additionally, the derivatives of the objective function are not usually available and must be obtained numerically if needed. Then, we restrict our search for a suitable optimization algorithm to metaheuristic gradientfree ones, testing Genetic Algorithm, Particle Swarm Optimization, Differential Evolution, Grey Wolf Optimization, Dragonfly Algorithm, and Whale Optimization Algorithm. Their performances on the considered problem were assessed and compared, in terms of accuracy and computational time.