Nonlinear ultrasonic nondestructive testing methods have largely improved early crack detection in solid materials. However, the nonlinear signature of the crack can be difficult to measure. In this paper, to overcome this drawback, we present a parametric optimization to automatically find the best driving frequency which maximizes the nonlinear signature. The optimization process is carried out in two steps to maximize a cost function corresponding to the nonlinear-linear ratio (NLR). From the signal received, NLR assesses the energy of the second harmonic over the energy of the fundamental component. This optimization is achieved by the hybridization of two algorithms: the genetic algorithm and the Nelder-Mead algorithm. The method is applied experimentally on a duraluminium sample including a crack. The optimal frequency is obtained after less than 14 iterations only. Moreover, the gain of the nonlinear-linear ratio reaches up to 12 dB in comparison with that obtained by the usual frequency setting