We propose as aim of this thesis to simultaneously schedule production tasks and machine maintenance activities in order to optimize both the storage costs and the delay penalties according to a Just In Time (JIT) approach. Often three classical research fields are investigated for this type of problems. The first is the complexity analysis of solving algorithms. The second is the formulation of algorithms giving the optimal solution. If the computational time is exorbitant, the third field could be investigated. It consists in using heuristics, meta-heuristics or the integration of different resolution methods to give approximate solutions. We will extend these fields to new algorithms based on deep learning and reinforcement learning. We favor learning-based algorithms that can effectively integrate optimization methods to find an efficient tradeoff between computational time and solution quality. However, the integrated schemes of learning and optimization methods are not obvious and easy to design and develop. Finally, the open questions that we need to answer are for example: which integration and hybridization scheme for which configuration of the studied production system? We also intend to improve the existing solving methods in terms of computational time and solution quality.