Industrial processes digitalization is nowadays becoming a trend. It permits a better analysis of their behavior and therefore can be used as a powerful management tool. This digitalization can be carried out in the shape of digital twins, which are, beyond a faithful three-dimensional representation of their elements, accurate digital models of the different processes or subprocess, characterizing the operation or behavior that the original system would have. Additionally, this modelling can be performed by means of machine learning techniques, creating in this way accurate models through the existing historical datasets of the original assets to be twinned. Following this goal, in the GEDIAV-H2O 2.0 project, two wastewater treatment plants have been modelled using different machine learning techniques. Both digital twins can be useful not only to compare the normal operation of the process and detect anomalies, but also to analyze the behavior in response to different external stimuli. The case study rises that the models allow the operators to predict the expected behaviour of the process variables in the plant, and can be also used for scenario simulation, being able to evaluate the effects of different configurations on the system without the real system being affected.