Modeling the dynamics of a servo mechanical system is one of the main challenges in robot manipulators, due to the the presence of complex and non-linear aspects of the phenomenons involved in it s modelling making it challenging to conceive an ideal representation. In this work we describe an original experimental bench composed of an electric actuator and a link joined by a joint. Using experimental measurements this projet has the objective of finding, through two different identification methods, a model that best represents the dynamics of the developed servosystem through experimental measurements. The first method consists in optimizing the parameters from different friction models using the grey-box type identification method. This work implements the Coulomb, Dahl and LuGre friction models. The blackbox method, utilizes artificial neural networks to predict the angular position and angular speed of the experimental bench. Analising the results, we observe that the models that considers the significant number of friction phenomena perform better. The results of the second method indicate a better performance than the grey-box method, showing a 67% reduction of the MAE error metric.