Ultrasonic non-destructive testing (NDT) controls the integrity of critical specimens. The multitude of tuning parameters for designing an ultrasonic NDT makes this task complex. Especially, phased array techniques (array of independently droved elements) improve performances of inspections but increase the number of parameters. Many methods and experimental feedback help to define parameters directly, but only for canonical cases. For non-canonical cases, expert have to perform a parametrical study with simulation tools which is long and fussy. In this communication, a method is proposed to help the expert whatever the configuration is, for applications such as positioning, tuning and sampling the transducer. To solve these optimization problems, we use an evolutionary algorithm called differential evolution (DE) that makes a population of individuals converge to the optimum. An individual is a candidate solution of the multivariable problem. The objective function is meaningful of the individual ability to solve the problem. The evaluation of the objective function is done with the NDT simulation platform CIVA developed by CEA LIST. The purpose is to propose a tool for NDT experts, in order to take care of the number of optimization's tuning parameters. To this aim, a special attention is paid to Randomize adaptive Differential Evolution (RaDE) proposed by Nobakhti and Wang which auto-adapt one optimization parameter. This algorithm has been tested on several NDT problems. In this communication, one or two examples of application are presented with the aim of determining the optimal transducer's configuration in order to detect defects inside a complex specimen