Taking optimal maintenance decisions is a challenging process as different maintenance actions have different effects on the system. Maintenance is defined as a set of associated techniques, tools and management actions that aim to maintain or restore the functioning state of the system. Maintenance excellence is the balance between performance and risk, therefore this helps improve the sustainability of production. Traditionally, maintenance decisions are taken based on human experience and on the basic information known about the system. With the availability of data collected during the system’s life cycle; machine learning approaches can help develop optimal strategies for maintenance actions. This paper proposes and depicts an optimization model imported from the machine learning field and developed to find optimal preventive maintenance strategies. The main objective of this developed optimization model is to minimize the downtime and allow the system to take autonomous decisions. In this work, the maintenance strategy is modeled as a Markov Decision Process (MDP). MDP is a classical forming of sequential decision-making problem.  Reinforcement learning (RL) model is then developed to solve the problem interactively. RL uses the MDP to define the interaction between the learning agent and the environment. The final output from this method in an optimal policy allows providing optimal actions in different situations.