Physical models of fluid power systems rely on the validity of the principles used for creating such models. In many cases, pump and motor performance is considered a large contributor to the efficiency of a whole fluid power system and, is used to approximate the behavior of the component and the system coupled to it. Often, estimates of the power losses and efficiency of pumps and motors is limited to manufacturer test data or simplified assumptions based on first principles. However, the use of the limited test data or idealized assumptions reduces the accuracy of the models and limits the validity of the theoretical results. Moreover, the creation of accurate physical models, their numerical implementation using a computer to solve the model and the experimental validation is time consuming and costly. New advances in machine learning, statistical analysis and numerical methods can be used to reduce the time used to develop a model of a pump or motor producing similar or better results. This paper proposes the use of an autonomous and iterative algorithm to obtain linear regression coefficients necessary to characterize the flow response of a pump or motor from existing experimental data. In this study a multivariate linear model for predicting the flow output of a pump or a motor is derived from experimental data by iteratively adding data points and by iteratively and autonomously testing regressor combinations to find the best possible flow model.