Observing processes as Discrete Event Systems (DESs) allows for the construction of formal models that assist analysis, control, optimization, correction, etc. Process Mining (PM) can be used to discover DES models from event logs. To support DES control, such discovered models must accurately capture the reachable state space and the patterns formed by event transitions. The diversity and complexity of the possible paths during PM introduce a trade-off between generalizing a model and preserving non-generalizable details that may be helpful, from the control perspective. This paper proposes a method to discover automata models particularly tailored for DES control. Based on premises that fit many of the systems discussed, our method is able to identify cycles that represent different complete tasks performed by the system. Furthermore, by formalizing the notions of fixed and variable repetition structures, we show how to capture recurring patterns in event logs and use them to guide the construction of models that balance precision and generalization. A comparison between the proposed approach and other state-of-the-art approaches usually employed in PM is also presented. It evidences the current gaps in the state-of-the-art and how our method addresses them. Case studies are used to illustrate our results.